2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-dpif.h"
20 #include "ofproto/ofproto-provider.h"
27 #include "byte-order.h"
28 #include "connectivity.h"
33 #include "dynamic-string.h"
34 #include "fail-open.h"
35 #include "guarded-list.h"
39 #include "mac-learning.h"
40 #include "meta-flow.h"
41 #include "multipath.h"
42 #include "netdev-vport.h"
47 #include "odp-execute.h"
50 #include "ofp-actions.h"
51 #include "ofp-parse.h"
52 #include "ofp-print.h"
53 #include "ofproto-dpif-ipfix.h"
54 #include "ofproto-dpif-mirror.h"
55 #include "ofproto-dpif-monitor.h"
56 #include "ofproto-dpif-rid.h"
57 #include "ofproto-dpif-sflow.h"
58 #include "ofproto-dpif-upcall.h"
59 #include "ofproto-dpif-xlate.h"
60 #include "poll-loop.h"
66 #include "unaligned.h"
68 #include "vlan-bitmap.h"
71 VLOG_DEFINE_THIS_MODULE(ofproto_dpif
);
73 COVERAGE_DEFINE(ofproto_dpif_expired
);
74 COVERAGE_DEFINE(packet_in_overflow
);
76 /* No bfd/cfm status change. */
77 #define NO_STATUS_CHANGE -1
86 * - Do include packets and bytes from datapath flows which have not
87 * recently been processed by a revalidator. */
88 struct ovs_mutex stats_mutex
;
89 struct dpif_flow_stats stats OVS_GUARDED
;
92 /* RULE_CAST() depends on this. */
93 BUILD_ASSERT_DECL(offsetof(struct rule_dpif
, up
) == 0);
95 static void rule_get_stats(struct rule
*, uint64_t *packets
, uint64_t *bytes
,
97 static struct rule_dpif
*rule_dpif_cast(const struct rule
*);
98 static void rule_expire(struct rule_dpif
*);
105 * - Do include packets and bytes from datapath flows which have not
106 * recently been processed by a revalidator. */
107 struct ovs_mutex stats_mutex
;
108 uint64_t packet_count OVS_GUARDED
; /* Number of packets received. */
109 uint64_t byte_count OVS_GUARDED
; /* Number of bytes received. */
110 struct bucket_counter
*bucket_stats OVS_GUARDED
; /* Bucket statistics. */
114 struct hmap_node hmap_node
; /* In struct ofproto's "bundles" hmap. */
115 struct ofproto_dpif
*ofproto
; /* Owning ofproto. */
116 void *aux
; /* Key supplied by ofproto's client. */
117 char *name
; /* Identifier for log messages. */
120 struct list ports
; /* Contains "struct ofport"s. */
121 enum port_vlan_mode vlan_mode
; /* VLAN mode */
122 int vlan
; /* -1=trunk port, else a 12-bit VLAN ID. */
123 unsigned long *trunks
; /* Bitmap of trunked VLANs, if 'vlan' == -1.
124 * NULL if all VLANs are trunked. */
125 struct lacp
*lacp
; /* LACP if LACP is enabled, otherwise NULL. */
126 struct bond
*bond
; /* Nonnull iff more than one port. */
127 bool use_priority_tags
; /* Use 802.1p tag for frames in VLAN 0? */
130 bool floodable
; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
133 static void bundle_remove(struct ofport
*);
134 static void bundle_update(struct ofbundle
*);
135 static void bundle_destroy(struct ofbundle
*);
136 static void bundle_del_port(struct ofport_dpif
*);
137 static void bundle_run(struct ofbundle
*);
138 static void bundle_wait(struct ofbundle
*);
140 static void stp_run(struct ofproto_dpif
*ofproto
);
141 static void stp_wait(struct ofproto_dpif
*ofproto
);
142 static int set_stp_port(struct ofport
*,
143 const struct ofproto_port_stp_settings
*);
146 struct hmap_node odp_port_node
; /* In dpif_backer's "odp_to_ofport_map". */
150 struct ofbundle
*bundle
; /* Bundle that contains this port, if any. */
151 struct list bundle_node
; /* In struct ofbundle's "ports" list. */
152 struct cfm
*cfm
; /* Connectivity Fault Management, if any. */
153 struct bfd
*bfd
; /* BFD, if any. */
154 bool may_enable
; /* May be enabled in bonds. */
155 bool is_tunnel
; /* This port is a tunnel. */
156 bool is_layer3
; /* This is a layer 3 port. */
157 long long int carrier_seq
; /* Carrier status changes. */
158 struct ofport_dpif
*peer
; /* Peer if patch port. */
161 struct stp_port
*stp_port
; /* Spanning Tree Protocol, if any. */
162 enum stp_state stp_state
; /* Always STP_DISABLED if STP not in use. */
163 long long int stp_state_entered
;
165 /* Queue to DSCP mapping. */
166 struct ofproto_port_queue
*qdscp
;
169 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
171 * This is deprecated. It is only for compatibility with broken device
172 * drivers in old versions of Linux that do not properly support VLANs when
173 * VLAN devices are not used. When broken device drivers are no longer in
174 * widespread use, we will delete these interfaces. */
175 ofp_port_t realdev_ofp_port
;
179 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
181 * This is deprecated. It is only for compatibility with broken device drivers
182 * in old versions of Linux that do not properly support VLANs when VLAN
183 * devices are not used. When broken device drivers are no longer in
184 * widespread use, we will delete these interfaces. */
185 struct vlan_splinter
{
186 struct hmap_node realdev_vid_node
;
187 struct hmap_node vlandev_node
;
188 ofp_port_t realdev_ofp_port
;
189 ofp_port_t vlandev_ofp_port
;
193 static void vsp_remove(struct ofport_dpif
*);
194 static void vsp_add(struct ofport_dpif
*, ofp_port_t realdev_ofp_port
, int vid
);
196 static odp_port_t
ofp_port_to_odp_port(const struct ofproto_dpif
*,
199 static ofp_port_t
odp_port_to_ofp_port(const struct ofproto_dpif
*,
202 static struct ofport_dpif
*
203 ofport_dpif_cast(const struct ofport
*ofport
)
205 return ofport
? CONTAINER_OF(ofport
, struct ofport_dpif
, up
) : NULL
;
208 static void port_run(struct ofport_dpif
*);
209 static int set_bfd(struct ofport
*, const struct smap
*);
210 static int set_cfm(struct ofport
*, const struct cfm_settings
*);
211 static void ofport_update_peer(struct ofport_dpif
*);
213 struct dpif_completion
{
214 struct list list_node
;
215 struct ofoperation
*op
;
218 /* Reasons that we might need to revalidate every datapath flow, and
219 * corresponding coverage counters.
221 * A value of 0 means that there is no need to revalidate.
223 * It would be nice to have some cleaner way to integrate with coverage
224 * counters, but with only a few reasons I guess this is good enough for
226 enum revalidate_reason
{
227 REV_RECONFIGURE
= 1, /* Switch configuration changed. */
228 REV_STP
, /* Spanning tree protocol port status change. */
229 REV_BOND
, /* Bonding changed. */
230 REV_PORT_TOGGLED
, /* Port enabled or disabled by CFM, LACP, ...*/
231 REV_FLOW_TABLE
, /* Flow table changed. */
232 REV_MAC_LEARNING
, /* Mac learning changed. */
234 COVERAGE_DEFINE(rev_reconfigure
);
235 COVERAGE_DEFINE(rev_stp
);
236 COVERAGE_DEFINE(rev_bond
);
237 COVERAGE_DEFINE(rev_port_toggled
);
238 COVERAGE_DEFINE(rev_flow_table
);
239 COVERAGE_DEFINE(rev_mac_learning
);
241 /* All datapaths of a given type share a single dpif backer instance. */
248 struct ovs_rwlock odp_to_ofport_lock
;
249 struct hmap odp_to_ofport_map OVS_GUARDED
; /* Contains "struct ofport"s. */
251 struct simap tnl_backers
; /* Set of dpif ports backing tunnels. */
253 enum revalidate_reason need_revalidate
; /* Revalidate all flows. */
255 bool recv_set_enable
; /* Enables or disables receiving packets. */
258 struct recirc_id_pool
*rid_pool
; /* Recirculation ID pool. */
259 bool enable_recirc
; /* True if the datapath supports recirculation */
261 /* True if the datapath supports variable-length
262 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions.
263 * False if the datapath supports only 8-byte (or shorter) userdata. */
264 bool variable_length_userdata
;
266 /* Maximum number of MPLS label stack entries that the datapath supports
268 size_t max_mpls_depth
;
271 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
272 static struct shash all_dpif_backers
= SHASH_INITIALIZER(&all_dpif_backers
);
274 struct ofproto_dpif
{
275 struct hmap_node all_ofproto_dpifs_node
; /* In 'all_ofproto_dpifs'. */
277 struct dpif_backer
*backer
;
279 uint64_t dump_seq
; /* Last read of udpif_dump_seq(). */
281 /* Special OpenFlow rules. */
282 struct rule_dpif
*miss_rule
; /* Sends flow table misses to controller. */
283 struct rule_dpif
*no_packet_in_rule
; /* Drops flow table misses. */
284 struct rule_dpif
*drop_frags_rule
; /* Used in OFPC_FRAG_DROP mode. */
287 struct netflow
*netflow
;
288 struct dpif_sflow
*sflow
;
289 struct dpif_ipfix
*ipfix
;
290 struct hmap bundles
; /* Contains "struct ofbundle"s. */
291 struct mac_learning
*ml
;
292 bool has_bonded_bundles
;
294 struct mbridge
*mbridge
;
296 struct ovs_mutex stats_mutex
;
297 struct netdev_stats stats OVS_GUARDED
; /* To account packets generated and
298 * consumed in userspace. */
302 long long int stp_last_tick
;
304 /* VLAN splinters. */
305 struct ovs_mutex vsp_mutex
;
306 struct hmap realdev_vid_map OVS_GUARDED
; /* (realdev,vid) -> vlandev. */
307 struct hmap vlandev_map OVS_GUARDED
; /* vlandev -> (realdev,vid). */
310 struct sset ports
; /* Set of standard port names. */
311 struct sset ghost_ports
; /* Ports with no datapath port. */
312 struct sset port_poll_set
; /* Queued names for port_poll() reply. */
313 int port_poll_errno
; /* Last errno for port_poll() reply. */
314 uint64_t change_seq
; /* Connectivity status changes. */
317 struct guarded_list pins
; /* Contains "struct ofputil_packet_in"s. */
318 struct seq
*pins_seq
; /* For notifying 'pins' reception. */
322 /* All existing ofproto_dpif instances, indexed by ->up.name. */
323 static struct hmap all_ofproto_dpifs
= HMAP_INITIALIZER(&all_ofproto_dpifs
);
325 static void ofproto_dpif_unixctl_init(void);
327 static inline struct ofproto_dpif
*
328 ofproto_dpif_cast(const struct ofproto
*ofproto
)
330 ovs_assert(ofproto
->ofproto_class
== &ofproto_dpif_class
);
331 return CONTAINER_OF(ofproto
, struct ofproto_dpif
, up
);
335 ofproto_dpif_get_max_mpls_depth(const struct ofproto_dpif
*ofproto
)
337 return ofproto
->backer
->max_mpls_depth
;
341 ofproto_dpif_get_enable_recirc(const struct ofproto_dpif
*ofproto
)
343 return ofproto
->backer
->enable_recirc
;
346 static struct ofport_dpif
*get_ofp_port(const struct ofproto_dpif
*ofproto
,
347 ofp_port_t ofp_port
);
348 static void ofproto_trace(struct ofproto_dpif
*, struct flow
*,
349 const struct ofpbuf
*packet
,
350 const struct ofpact
[], size_t ofpacts_len
,
353 /* Global variables. */
354 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
356 /* Initial mappings of port to bridge mappings. */
357 static struct shash init_ofp_ports
= SHASH_INITIALIZER(&init_ofp_ports
);
359 /* Executes 'fm'. The caller retains ownership of 'fm' and everything in
362 ofproto_dpif_flow_mod(struct ofproto_dpif
*ofproto
,
363 struct ofputil_flow_mod
*fm
)
365 ofproto_flow_mod(&ofproto
->up
, fm
);
368 /* Resets the modified time for 'rule' or an equivalent rule. If 'rule' is not
369 * in the classifier, but an equivalent rule is, unref 'rule' and ref the new
370 * rule. Otherwise if 'rule' is no longer installed in the classifier,
373 * Returns the rule whose modified time has been reset. */
375 ofproto_dpif_refresh_rule(struct rule_dpif
*rule
)
377 return rule_dpif_cast(ofproto_refresh_rule(&rule
->up
));
380 /* Appends 'pin' to the queue of "packet ins" to be sent to the controller.
381 * Takes ownership of 'pin' and pin->packet. */
383 ofproto_dpif_send_packet_in(struct ofproto_dpif
*ofproto
,
384 struct ofproto_packet_in
*pin
)
386 if (!guarded_list_push_back(&ofproto
->pins
, &pin
->list_node
, 1024)) {
387 COVERAGE_INC(packet_in_overflow
);
388 free(CONST_CAST(void *, pin
->up
.packet
));
392 /* Wakes up main thread for packet-in I/O. */
393 seq_change(ofproto
->pins_seq
);
396 /* The default "table-miss" behaviour for OpenFlow1.3+ is to drop the
397 * packet rather than to send the packet to the controller.
399 * This function returns false to indicate that a packet_in message
400 * for a "table-miss" should be sent to at least one controller.
401 * False otherwise. */
403 ofproto_dpif_wants_packet_in_on_miss(struct ofproto_dpif
*ofproto
)
405 return connmgr_wants_packet_in_on_miss(ofproto
->up
.connmgr
);
408 /* Factory functions. */
411 init(const struct shash
*iface_hints
)
413 struct shash_node
*node
;
415 /* Make a local copy, since we don't own 'iface_hints' elements. */
416 SHASH_FOR_EACH(node
, iface_hints
) {
417 const struct iface_hint
*orig_hint
= node
->data
;
418 struct iface_hint
*new_hint
= xmalloc(sizeof *new_hint
);
420 new_hint
->br_name
= xstrdup(orig_hint
->br_name
);
421 new_hint
->br_type
= xstrdup(orig_hint
->br_type
);
422 new_hint
->ofp_port
= orig_hint
->ofp_port
;
424 shash_add(&init_ofp_ports
, node
->name
, new_hint
);
429 enumerate_types(struct sset
*types
)
431 dp_enumerate_types(types
);
435 enumerate_names(const char *type
, struct sset
*names
)
437 struct ofproto_dpif
*ofproto
;
440 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
441 if (strcmp(type
, ofproto
->up
.type
)) {
444 sset_add(names
, ofproto
->up
.name
);
451 del(const char *type
, const char *name
)
456 error
= dpif_open(name
, type
, &dpif
);
458 error
= dpif_delete(dpif
);
465 port_open_type(const char *datapath_type
, const char *port_type
)
467 return dpif_port_open_type(datapath_type
, port_type
);
470 /* Type functions. */
472 static void process_dpif_port_changes(struct dpif_backer
*);
473 static void process_dpif_all_ports_changed(struct dpif_backer
*);
474 static void process_dpif_port_change(struct dpif_backer
*,
475 const char *devname
);
476 static void process_dpif_port_error(struct dpif_backer
*, int error
);
478 static struct ofproto_dpif
*
479 lookup_ofproto_dpif_by_port_name(const char *name
)
481 struct ofproto_dpif
*ofproto
;
483 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
484 if (sset_contains(&ofproto
->ports
, name
)) {
493 type_run(const char *type
)
495 struct dpif_backer
*backer
;
497 backer
= shash_find_data(&all_dpif_backers
, type
);
499 /* This is not necessarily a problem, since backers are only
500 * created on demand. */
504 dpif_run(backer
->dpif
);
506 /* If vswitchd started with other_config:flow_restore_wait set as "true",
507 * and the configuration has now changed to "false", enable receiving
508 * packets from the datapath. */
509 if (!backer
->recv_set_enable
&& !ofproto_get_flow_restore_wait()) {
512 backer
->recv_set_enable
= true;
514 error
= dpif_recv_set(backer
->dpif
, backer
->recv_set_enable
);
516 VLOG_ERR("Failed to enable receiving packets in dpif.");
519 dpif_flow_flush(backer
->dpif
);
520 backer
->need_revalidate
= REV_RECONFIGURE
;
523 if (backer
->recv_set_enable
) {
524 udpif_set_threads(backer
->udpif
, n_handlers
, n_revalidators
);
527 if (backer
->need_revalidate
) {
528 struct ofproto_dpif
*ofproto
;
529 struct simap_node
*node
;
530 struct simap tmp_backers
;
532 /* Handle tunnel garbage collection. */
533 simap_init(&tmp_backers
);
534 simap_swap(&backer
->tnl_backers
, &tmp_backers
);
536 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
537 struct ofport_dpif
*iter
;
539 if (backer
!= ofproto
->backer
) {
543 HMAP_FOR_EACH (iter
, up
.hmap_node
, &ofproto
->up
.ports
) {
544 char namebuf
[NETDEV_VPORT_NAME_BUFSIZE
];
547 if (!iter
->is_tunnel
) {
551 dp_port
= netdev_vport_get_dpif_port(iter
->up
.netdev
,
552 namebuf
, sizeof namebuf
);
553 node
= simap_find(&tmp_backers
, dp_port
);
555 simap_put(&backer
->tnl_backers
, dp_port
, node
->data
);
556 simap_delete(&tmp_backers
, node
);
557 node
= simap_find(&backer
->tnl_backers
, dp_port
);
559 node
= simap_find(&backer
->tnl_backers
, dp_port
);
561 odp_port_t odp_port
= ODPP_NONE
;
563 if (!dpif_port_add(backer
->dpif
, iter
->up
.netdev
,
565 simap_put(&backer
->tnl_backers
, dp_port
,
566 odp_to_u32(odp_port
));
567 node
= simap_find(&backer
->tnl_backers
, dp_port
);
572 iter
->odp_port
= node
? u32_to_odp(node
->data
) : ODPP_NONE
;
573 if (tnl_port_reconfigure(iter
, iter
->up
.netdev
,
575 backer
->need_revalidate
= REV_RECONFIGURE
;
580 SIMAP_FOR_EACH (node
, &tmp_backers
) {
581 dpif_port_del(backer
->dpif
, u32_to_odp(node
->data
));
583 simap_destroy(&tmp_backers
);
585 switch (backer
->need_revalidate
) {
586 case REV_RECONFIGURE
: COVERAGE_INC(rev_reconfigure
); break;
587 case REV_STP
: COVERAGE_INC(rev_stp
); break;
588 case REV_BOND
: COVERAGE_INC(rev_bond
); break;
589 case REV_PORT_TOGGLED
: COVERAGE_INC(rev_port_toggled
); break;
590 case REV_FLOW_TABLE
: COVERAGE_INC(rev_flow_table
); break;
591 case REV_MAC_LEARNING
: COVERAGE_INC(rev_mac_learning
); break;
593 backer
->need_revalidate
= 0;
595 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
596 struct ofport_dpif
*ofport
;
597 struct ofbundle
*bundle
;
599 if (ofproto
->backer
!= backer
) {
603 ovs_rwlock_wrlock(&xlate_rwlock
);
604 xlate_ofproto_set(ofproto
, ofproto
->up
.name
,
605 ofproto
->backer
->dpif
, ofproto
->miss_rule
,
606 ofproto
->no_packet_in_rule
, ofproto
->ml
,
607 ofproto
->stp
, ofproto
->mbridge
,
608 ofproto
->sflow
, ofproto
->ipfix
,
609 ofproto
->netflow
, ofproto
->up
.frag_handling
,
610 ofproto
->up
.forward_bpdu
,
611 connmgr_has_in_band(ofproto
->up
.connmgr
),
612 ofproto
->backer
->enable_recirc
,
613 ofproto
->backer
->variable_length_userdata
,
614 ofproto
->backer
->max_mpls_depth
);
616 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
617 xlate_bundle_set(ofproto
, bundle
, bundle
->name
,
618 bundle
->vlan_mode
, bundle
->vlan
,
619 bundle
->trunks
, bundle
->use_priority_tags
,
620 bundle
->bond
, bundle
->lacp
,
624 HMAP_FOR_EACH (ofport
, up
.hmap_node
, &ofproto
->up
.ports
) {
625 int stp_port
= ofport
->stp_port
626 ? stp_port_no(ofport
->stp_port
)
628 xlate_ofport_set(ofproto
, ofport
->bundle
, ofport
,
629 ofport
->up
.ofp_port
, ofport
->odp_port
,
630 ofport
->up
.netdev
, ofport
->cfm
,
631 ofport
->bfd
, ofport
->peer
, stp_port
,
632 ofport
->qdscp
, ofport
->n_qdscp
,
633 ofport
->up
.pp
.config
, ofport
->up
.pp
.state
,
634 ofport
->is_tunnel
, ofport
->may_enable
);
636 ovs_rwlock_unlock(&xlate_rwlock
);
639 udpif_revalidate(backer
->udpif
);
642 process_dpif_port_changes(backer
);
647 /* Check for and handle port changes in 'backer''s dpif. */
649 process_dpif_port_changes(struct dpif_backer
*backer
)
655 error
= dpif_port_poll(backer
->dpif
, &devname
);
661 process_dpif_all_ports_changed(backer
);
665 process_dpif_port_change(backer
, devname
);
670 process_dpif_port_error(backer
, error
);
677 process_dpif_all_ports_changed(struct dpif_backer
*backer
)
679 struct ofproto_dpif
*ofproto
;
680 struct dpif_port dpif_port
;
681 struct dpif_port_dump dump
;
682 struct sset devnames
;
685 sset_init(&devnames
);
686 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
687 if (ofproto
->backer
== backer
) {
688 struct ofport
*ofport
;
690 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->up
.ports
) {
691 sset_add(&devnames
, netdev_get_name(ofport
->netdev
));
695 DPIF_PORT_FOR_EACH (&dpif_port
, &dump
, backer
->dpif
) {
696 sset_add(&devnames
, dpif_port
.name
);
699 SSET_FOR_EACH (devname
, &devnames
) {
700 process_dpif_port_change(backer
, devname
);
702 sset_destroy(&devnames
);
706 process_dpif_port_change(struct dpif_backer
*backer
, const char *devname
)
708 struct ofproto_dpif
*ofproto
;
709 struct dpif_port port
;
711 /* Don't report on the datapath's device. */
712 if (!strcmp(devname
, dpif_base_name(backer
->dpif
))) {
716 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
,
717 &all_ofproto_dpifs
) {
718 if (simap_contains(&ofproto
->backer
->tnl_backers
, devname
)) {
723 ofproto
= lookup_ofproto_dpif_by_port_name(devname
);
724 if (dpif_port_query_by_name(backer
->dpif
, devname
, &port
)) {
725 /* The port was removed. If we know the datapath,
726 * report it through poll_set(). If we don't, it may be
727 * notifying us of a removal we initiated, so ignore it.
728 * If there's a pending ENOBUFS, let it stand, since
729 * everything will be reevaluated. */
730 if (ofproto
&& ofproto
->port_poll_errno
!= ENOBUFS
) {
731 sset_add(&ofproto
->port_poll_set
, devname
);
732 ofproto
->port_poll_errno
= 0;
734 } else if (!ofproto
) {
735 /* The port was added, but we don't know with which
736 * ofproto we should associate it. Delete it. */
737 dpif_port_del(backer
->dpif
, port
.port_no
);
739 struct ofport_dpif
*ofport
;
741 ofport
= ofport_dpif_cast(shash_find_data(
742 &ofproto
->up
.port_by_name
, devname
));
744 && ofport
->odp_port
!= port
.port_no
745 && !odp_port_to_ofport(backer
, port
.port_no
))
747 /* 'ofport''s datapath port number has changed from
748 * 'ofport->odp_port' to 'port.port_no'. Update our internal data
749 * structures to match. */
750 ovs_rwlock_wrlock(&backer
->odp_to_ofport_lock
);
751 hmap_remove(&backer
->odp_to_ofport_map
, &ofport
->odp_port_node
);
752 ofport
->odp_port
= port
.port_no
;
753 hmap_insert(&backer
->odp_to_ofport_map
, &ofport
->odp_port_node
,
754 hash_odp_port(port
.port_no
));
755 ovs_rwlock_unlock(&backer
->odp_to_ofport_lock
);
756 backer
->need_revalidate
= REV_RECONFIGURE
;
759 dpif_port_destroy(&port
);
762 /* Propagate 'error' to all ofprotos based on 'backer'. */
764 process_dpif_port_error(struct dpif_backer
*backer
, int error
)
766 struct ofproto_dpif
*ofproto
;
768 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
769 if (ofproto
->backer
== backer
) {
770 sset_clear(&ofproto
->port_poll_set
);
771 ofproto
->port_poll_errno
= error
;
777 type_wait(const char *type
)
779 struct dpif_backer
*backer
;
781 backer
= shash_find_data(&all_dpif_backers
, type
);
783 /* This is not necessarily a problem, since backers are only
784 * created on demand. */
788 dpif_wait(backer
->dpif
);
791 /* Basic life-cycle. */
793 static int add_internal_flows(struct ofproto_dpif
*);
795 static struct ofproto
*
798 struct ofproto_dpif
*ofproto
= xmalloc(sizeof *ofproto
);
803 dealloc(struct ofproto
*ofproto_
)
805 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
810 close_dpif_backer(struct dpif_backer
*backer
)
812 ovs_assert(backer
->refcount
> 0);
814 if (--backer
->refcount
) {
818 udpif_destroy(backer
->udpif
);
820 simap_destroy(&backer
->tnl_backers
);
821 ovs_rwlock_destroy(&backer
->odp_to_ofport_lock
);
822 hmap_destroy(&backer
->odp_to_ofport_map
);
823 shash_find_and_delete(&all_dpif_backers
, backer
->type
);
824 recirc_id_pool_destroy(backer
->rid_pool
);
826 dpif_close(backer
->dpif
);
830 /* Datapath port slated for removal from datapath. */
832 struct list list_node
;
836 static bool check_variable_length_userdata(struct dpif_backer
*backer
);
837 static size_t check_max_mpls_depth(struct dpif_backer
*backer
);
838 static bool check_recirc(struct dpif_backer
*backer
);
841 open_dpif_backer(const char *type
, struct dpif_backer
**backerp
)
843 struct dpif_backer
*backer
;
844 struct dpif_port_dump port_dump
;
845 struct dpif_port port
;
846 struct shash_node
*node
;
847 struct list garbage_list
;
848 struct odp_garbage
*garbage
, *next
;
855 backer
= shash_find_data(&all_dpif_backers
, type
);
862 backer_name
= xasprintf("ovs-%s", type
);
864 /* Remove any existing datapaths, since we assume we're the only
865 * userspace controlling the datapath. */
867 dp_enumerate_names(type
, &names
);
868 SSET_FOR_EACH(name
, &names
) {
869 struct dpif
*old_dpif
;
871 /* Don't remove our backer if it exists. */
872 if (!strcmp(name
, backer_name
)) {
876 if (dpif_open(name
, type
, &old_dpif
)) {
877 VLOG_WARN("couldn't open old datapath %s to remove it", name
);
879 dpif_delete(old_dpif
);
880 dpif_close(old_dpif
);
883 sset_destroy(&names
);
885 backer
= xmalloc(sizeof *backer
);
887 error
= dpif_create_and_open(backer_name
, type
, &backer
->dpif
);
890 VLOG_ERR("failed to open datapath of type %s: %s", type
,
891 ovs_strerror(error
));
895 backer
->udpif
= udpif_create(backer
, backer
->dpif
);
897 backer
->type
= xstrdup(type
);
898 backer
->refcount
= 1;
899 hmap_init(&backer
->odp_to_ofport_map
);
900 ovs_rwlock_init(&backer
->odp_to_ofport_lock
);
901 backer
->need_revalidate
= 0;
902 simap_init(&backer
->tnl_backers
);
903 backer
->recv_set_enable
= !ofproto_get_flow_restore_wait();
906 if (backer
->recv_set_enable
) {
907 dpif_flow_flush(backer
->dpif
);
910 /* Loop through the ports already on the datapath and remove any
911 * that we don't need anymore. */
912 list_init(&garbage_list
);
913 dpif_port_dump_start(&port_dump
, backer
->dpif
);
914 while (dpif_port_dump_next(&port_dump
, &port
)) {
915 node
= shash_find(&init_ofp_ports
, port
.name
);
916 if (!node
&& strcmp(port
.name
, dpif_base_name(backer
->dpif
))) {
917 garbage
= xmalloc(sizeof *garbage
);
918 garbage
->odp_port
= port
.port_no
;
919 list_push_front(&garbage_list
, &garbage
->list_node
);
922 dpif_port_dump_done(&port_dump
);
924 LIST_FOR_EACH_SAFE (garbage
, next
, list_node
, &garbage_list
) {
925 dpif_port_del(backer
->dpif
, garbage
->odp_port
);
926 list_remove(&garbage
->list_node
);
930 shash_add(&all_dpif_backers
, type
, backer
);
932 error
= dpif_recv_set(backer
->dpif
, backer
->recv_set_enable
);
934 VLOG_ERR("failed to listen on datapath of type %s: %s",
935 type
, ovs_strerror(error
));
936 close_dpif_backer(backer
);
939 backer
->enable_recirc
= check_recirc(backer
);
940 backer
->variable_length_userdata
= check_variable_length_userdata(backer
);
941 backer
->max_mpls_depth
= check_max_mpls_depth(backer
);
942 backer
->rid_pool
= recirc_id_pool_create();
944 if (backer
->recv_set_enable
) {
945 udpif_set_threads(backer
->udpif
, n_handlers
, n_revalidators
);
951 /* Tests whether 'backer''s datapath supports recirculation Only newer datapath
952 * supports OVS_KEY_ATTR in OVS_ACTION_ATTR_USERSPACE actions. We need to
953 * disable some features on older datapaths that don't support this feature.
955 * Returns false if 'backer' definitely does not support recirculation, true if
956 * it seems to support recirculation or if at least the error we get is
959 check_recirc(struct dpif_backer
*backer
)
962 struct odputil_keybuf keybuf
;
965 bool enable_recirc
= false;
967 memset(&flow
, 0, sizeof flow
);
971 ofpbuf_use_stack(&key
, &keybuf
, sizeof keybuf
);
972 odp_flow_key_from_flow(&key
, &flow
, NULL
, 0);
974 error
= dpif_flow_put(backer
->dpif
, DPIF_FP_CREATE
| DPIF_FP_MODIFY
,
975 ofpbuf_data(&key
), ofpbuf_size(&key
), NULL
, 0, NULL
,
977 if (error
&& error
!= EEXIST
) {
978 if (error
!= EINVAL
) {
979 VLOG_WARN("%s: Reciculation flow probe failed (%s)",
980 dpif_name(backer
->dpif
), ovs_strerror(error
));
985 error
= dpif_flow_del(backer
->dpif
, ofpbuf_data(&key
), ofpbuf_size(&key
),
988 VLOG_WARN("%s: failed to delete recirculation feature probe flow",
989 dpif_name(backer
->dpif
));
992 enable_recirc
= true;
996 VLOG_INFO("%s: Datapath supports recirculation",
997 dpif_name(backer
->dpif
));
999 VLOG_INFO("%s: Datapath does not support recirculation",
1000 dpif_name(backer
->dpif
));
1003 return enable_recirc
;
1006 /* Tests whether 'backer''s datapath supports variable-length
1007 * OVS_USERSPACE_ATTR_USERDATA in OVS_ACTION_ATTR_USERSPACE actions. We need
1008 * to disable some features on older datapaths that don't support this
1011 * Returns false if 'backer' definitely does not support variable-length
1012 * userdata, true if it seems to support them or if at least the error we get
1015 check_variable_length_userdata(struct dpif_backer
*backer
)
1017 struct eth_header
*eth
;
1018 struct ofpbuf actions
;
1019 struct dpif_execute execute
;
1020 struct ofpbuf packet
;
1024 /* Compose a userspace action that will cause an ERANGE error on older
1025 * datapaths that don't support variable-length userdata.
1027 * We really test for using userdata longer than 8 bytes, but older
1028 * datapaths accepted these, silently truncating the userdata to 8 bytes.
1029 * The same older datapaths rejected userdata shorter than 8 bytes, so we
1030 * test for that instead as a proxy for longer userdata support. */
1031 ofpbuf_init(&actions
, 64);
1032 start
= nl_msg_start_nested(&actions
, OVS_ACTION_ATTR_USERSPACE
);
1033 nl_msg_put_u32(&actions
, OVS_USERSPACE_ATTR_PID
,
1034 dpif_port_get_pid(backer
->dpif
, ODPP_NONE
, 0));
1035 nl_msg_put_unspec_zero(&actions
, OVS_USERSPACE_ATTR_USERDATA
, 4);
1036 nl_msg_end_nested(&actions
, start
);
1038 /* Compose a dummy ethernet packet. */
1039 ofpbuf_init(&packet
, ETH_HEADER_LEN
);
1040 eth
= ofpbuf_put_zeros(&packet
, ETH_HEADER_LEN
);
1041 eth
->eth_type
= htons(0x1234);
1043 /* Execute the actions. On older datapaths this fails with ERANGE, on
1044 * newer datapaths it succeeds. */
1045 execute
.actions
= ofpbuf_data(&actions
);
1046 execute
.actions_len
= ofpbuf_size(&actions
);
1047 execute
.packet
= &packet
;
1048 execute
.md
= PKT_METADATA_INITIALIZER(0);
1049 execute
.needs_help
= false;
1051 error
= dpif_execute(backer
->dpif
, &execute
);
1053 ofpbuf_uninit(&packet
);
1054 ofpbuf_uninit(&actions
);
1058 /* Variable-length userdata is supported.
1060 * Purge received packets to avoid processing the nonsense packet we
1061 * sent to userspace, then report success. */
1062 dpif_recv_purge(backer
->dpif
);
1066 /* Variable-length userdata is not supported. */
1067 VLOG_WARN("%s: datapath does not support variable-length userdata "
1068 "feature (needs Linux 3.10+ or kernel module from OVS "
1069 "1..11+). The NXAST_SAMPLE action will be ignored.",
1070 dpif_name(backer
->dpif
));
1074 /* Something odd happened. We're not sure whether variable-length
1075 * userdata is supported. Default to "yes". */
1076 VLOG_WARN("%s: variable-length userdata feature probe failed (%s)",
1077 dpif_name(backer
->dpif
), ovs_strerror(error
));
1082 /* Tests the MPLS label stack depth supported by 'backer''s datapath.
1084 * Returns the number of elements in a struct flow's mpls_lse field
1085 * if the datapath supports at least that many entries in an
1087 * Otherwise returns the number of MPLS push actions supported by
1090 check_max_mpls_depth(struct dpif_backer
*backer
)
1095 for (n
= 0; n
< FLOW_MAX_MPLS_LABELS
; n
++) {
1096 struct odputil_keybuf keybuf
;
1100 memset(&flow
, 0, sizeof flow
);
1101 flow
.dl_type
= htons(ETH_TYPE_MPLS
);
1102 flow_set_mpls_bos(&flow
, n
, 1);
1104 ofpbuf_use_stack(&key
, &keybuf
, sizeof keybuf
);
1105 odp_flow_key_from_flow(&key
, &flow
, NULL
, 0);
1107 error
= dpif_flow_put(backer
->dpif
, DPIF_FP_CREATE
| DPIF_FP_MODIFY
,
1108 ofpbuf_data(&key
), ofpbuf_size(&key
), NULL
, 0, NULL
, 0, NULL
);
1109 if (error
&& error
!= EEXIST
) {
1110 if (error
!= EINVAL
) {
1111 VLOG_WARN("%s: MPLS stack length feature probe failed (%s)",
1112 dpif_name(backer
->dpif
), ovs_strerror(error
));
1117 error
= dpif_flow_del(backer
->dpif
, ofpbuf_data(&key
), ofpbuf_size(&key
), NULL
);
1119 VLOG_WARN("%s: failed to delete MPLS feature probe flow",
1120 dpif_name(backer
->dpif
));
1124 VLOG_INFO("%s: MPLS label stack length probed as %d",
1125 dpif_name(backer
->dpif
), n
);
1130 construct(struct ofproto
*ofproto_
)
1132 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1133 struct shash_node
*node
, *next
;
1136 error
= open_dpif_backer(ofproto
->up
.type
, &ofproto
->backer
);
1141 ofproto
->netflow
= NULL
;
1142 ofproto
->sflow
= NULL
;
1143 ofproto
->ipfix
= NULL
;
1144 ofproto
->stp
= NULL
;
1145 ofproto
->dump_seq
= 0;
1146 hmap_init(&ofproto
->bundles
);
1147 ofproto
->ml
= mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME
);
1148 ofproto
->mbridge
= mbridge_create();
1149 ofproto
->has_bonded_bundles
= false;
1150 ofproto
->lacp_enabled
= false;
1151 ovs_mutex_init_adaptive(&ofproto
->stats_mutex
);
1152 ovs_mutex_init(&ofproto
->vsp_mutex
);
1154 guarded_list_init(&ofproto
->pins
);
1156 ofproto_dpif_unixctl_init();
1158 hmap_init(&ofproto
->vlandev_map
);
1159 hmap_init(&ofproto
->realdev_vid_map
);
1161 sset_init(&ofproto
->ports
);
1162 sset_init(&ofproto
->ghost_ports
);
1163 sset_init(&ofproto
->port_poll_set
);
1164 ofproto
->port_poll_errno
= 0;
1165 ofproto
->change_seq
= 0;
1166 ofproto
->pins_seq
= seq_create();
1167 ofproto
->pins_seqno
= seq_read(ofproto
->pins_seq
);
1170 SHASH_FOR_EACH_SAFE (node
, next
, &init_ofp_ports
) {
1171 struct iface_hint
*iface_hint
= node
->data
;
1173 if (!strcmp(iface_hint
->br_name
, ofproto
->up
.name
)) {
1174 /* Check if the datapath already has this port. */
1175 if (dpif_port_exists(ofproto
->backer
->dpif
, node
->name
)) {
1176 sset_add(&ofproto
->ports
, node
->name
);
1179 free(iface_hint
->br_name
);
1180 free(iface_hint
->br_type
);
1182 shash_delete(&init_ofp_ports
, node
);
1186 hmap_insert(&all_ofproto_dpifs
, &ofproto
->all_ofproto_dpifs_node
,
1187 hash_string(ofproto
->up
.name
, 0));
1188 memset(&ofproto
->stats
, 0, sizeof ofproto
->stats
);
1190 ofproto_init_tables(ofproto_
, N_TABLES
);
1191 error
= add_internal_flows(ofproto
);
1193 ofproto
->up
.tables
[TBL_INTERNAL
].flags
= OFTABLE_HIDDEN
| OFTABLE_READONLY
;
1199 add_internal_miss_flow(struct ofproto_dpif
*ofproto
, int id
,
1200 const struct ofpbuf
*ofpacts
, struct rule_dpif
**rulep
)
1206 match_init_catchall(&match
);
1207 match_set_reg(&match
, 0, id
);
1209 error
= ofproto_dpif_add_internal_flow(ofproto
, &match
, 0, ofpacts
, &rule
);
1210 *rulep
= error
? NULL
: rule_dpif_cast(rule
);
1216 add_internal_flows(struct ofproto_dpif
*ofproto
)
1218 struct ofpact_controller
*controller
;
1219 uint64_t ofpacts_stub
[128 / 8];
1220 struct ofpbuf ofpacts
;
1221 struct rule
*unused_rulep OVS_UNUSED
;
1222 struct ofpact_resubmit
*resubmit
;
1227 ofpbuf_use_stack(&ofpacts
, ofpacts_stub
, sizeof ofpacts_stub
);
1230 controller
= ofpact_put_CONTROLLER(&ofpacts
);
1231 controller
->max_len
= UINT16_MAX
;
1232 controller
->controller_id
= 0;
1233 controller
->reason
= OFPR_NO_MATCH
;
1234 ofpact_pad(&ofpacts
);
1236 error
= add_internal_miss_flow(ofproto
, id
++, &ofpacts
,
1237 &ofproto
->miss_rule
);
1242 ofpbuf_clear(&ofpacts
);
1243 error
= add_internal_miss_flow(ofproto
, id
++, &ofpacts
,
1244 &ofproto
->no_packet_in_rule
);
1249 error
= add_internal_miss_flow(ofproto
, id
++, &ofpacts
,
1250 &ofproto
->drop_frags_rule
);
1255 /* Continue non-recirculation rule lookups from table 0.
1257 * (priority=2), recirc=0, actions=resubmit(, 0)
1259 resubmit
= ofpact_put_RESUBMIT(&ofpacts
);
1260 resubmit
->ofpact
.compat
= 0;
1261 resubmit
->in_port
= OFPP_IN_PORT
;
1262 resubmit
->table_id
= 0;
1264 match_init_catchall(&match
);
1265 match_set_recirc_id(&match
, 0);
1267 error
= ofproto_dpif_add_internal_flow(ofproto
, &match
, 2, &ofpacts
,
1273 /* Drop any run away recirc rule lookups. Recirc_id has to be
1274 * non-zero when reaching this rule.
1276 * (priority=1), *, actions=drop
1278 ofpbuf_clear(&ofpacts
);
1279 match_init_catchall(&match
);
1280 error
= ofproto_dpif_add_internal_flow(ofproto
, &match
, 1, &ofpacts
,
1287 destruct(struct ofproto
*ofproto_
)
1289 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1290 struct rule_dpif
*rule
, *next_rule
;
1291 struct ofproto_packet_in
*pin
, *next_pin
;
1292 struct oftable
*table
;
1295 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1296 ovs_rwlock_wrlock(&xlate_rwlock
);
1297 xlate_remove_ofproto(ofproto
);
1298 ovs_rwlock_unlock(&xlate_rwlock
);
1300 /* Ensure that the upcall processing threads have no remaining references
1301 * to the ofproto or anything in it. */
1302 udpif_synchronize(ofproto
->backer
->udpif
);
1304 hmap_remove(&all_ofproto_dpifs
, &ofproto
->all_ofproto_dpifs_node
);
1306 OFPROTO_FOR_EACH_TABLE (table
, &ofproto
->up
) {
1307 struct cls_cursor cursor
;
1309 fat_rwlock_rdlock(&table
->cls
.rwlock
);
1310 cls_cursor_init(&cursor
, &table
->cls
, NULL
);
1311 fat_rwlock_unlock(&table
->cls
.rwlock
);
1312 CLS_CURSOR_FOR_EACH_SAFE (rule
, next_rule
, up
.cr
, &cursor
) {
1313 ofproto_rule_delete(&ofproto
->up
, &rule
->up
);
1317 guarded_list_pop_all(&ofproto
->pins
, &pins
);
1318 LIST_FOR_EACH_SAFE (pin
, next_pin
, list_node
, &pins
) {
1319 list_remove(&pin
->list_node
);
1320 free(CONST_CAST(void *, pin
->up
.packet
));
1323 guarded_list_destroy(&ofproto
->pins
);
1325 mbridge_unref(ofproto
->mbridge
);
1327 netflow_unref(ofproto
->netflow
);
1328 dpif_sflow_unref(ofproto
->sflow
);
1329 hmap_destroy(&ofproto
->bundles
);
1330 mac_learning_unref(ofproto
->ml
);
1332 hmap_destroy(&ofproto
->vlandev_map
);
1333 hmap_destroy(&ofproto
->realdev_vid_map
);
1335 sset_destroy(&ofproto
->ports
);
1336 sset_destroy(&ofproto
->ghost_ports
);
1337 sset_destroy(&ofproto
->port_poll_set
);
1339 ovs_mutex_destroy(&ofproto
->stats_mutex
);
1340 ovs_mutex_destroy(&ofproto
->vsp_mutex
);
1342 seq_destroy(ofproto
->pins_seq
);
1344 close_dpif_backer(ofproto
->backer
);
1348 run(struct ofproto
*ofproto_
)
1350 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1351 uint64_t new_seq
, new_dump_seq
;
1353 if (mbridge_need_revalidate(ofproto
->mbridge
)) {
1354 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1355 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
1356 mac_learning_flush(ofproto
->ml
);
1357 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
1360 /* Do not perform any periodic activity required by 'ofproto' while
1361 * waiting for flow restore to complete. */
1362 if (!ofproto_get_flow_restore_wait()) {
1363 struct ofproto_packet_in
*pin
, *next_pin
;
1366 guarded_list_pop_all(&ofproto
->pins
, &pins
);
1367 LIST_FOR_EACH_SAFE (pin
, next_pin
, list_node
, &pins
) {
1368 connmgr_send_packet_in(ofproto
->up
.connmgr
, pin
);
1369 list_remove(&pin
->list_node
);
1370 free(CONST_CAST(void *, pin
->up
.packet
));
1375 /* Always updates the ofproto->pins_seqno to avoid frequent wakeup during
1376 * flow restore. Even though nothing is processed during flow restore,
1377 * all queued 'pins' will be handled immediately when flow restore
1379 ofproto
->pins_seqno
= seq_read(ofproto
->pins_seq
);
1381 if (ofproto
->netflow
) {
1382 netflow_run(ofproto
->netflow
);
1384 if (ofproto
->sflow
) {
1385 dpif_sflow_run(ofproto
->sflow
);
1387 if (ofproto
->ipfix
) {
1388 dpif_ipfix_run(ofproto
->ipfix
);
1391 new_seq
= seq_read(connectivity_seq_get());
1392 if (ofproto
->change_seq
!= new_seq
) {
1393 struct ofport_dpif
*ofport
;
1395 HMAP_FOR_EACH (ofport
, up
.hmap_node
, &ofproto
->up
.ports
) {
1399 ofproto
->change_seq
= new_seq
;
1401 if (ofproto
->lacp_enabled
|| ofproto
->has_bonded_bundles
) {
1402 struct ofbundle
*bundle
;
1404 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
1410 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
1411 if (mac_learning_run(ofproto
->ml
)) {
1412 ofproto
->backer
->need_revalidate
= REV_MAC_LEARNING
;
1414 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
1416 new_dump_seq
= seq_read(udpif_dump_seq(ofproto
->backer
->udpif
));
1417 if (ofproto
->dump_seq
!= new_dump_seq
) {
1418 struct rule
*rule
, *next_rule
;
1420 /* We know stats are relatively fresh, so now is a good time to do some
1422 ofproto
->dump_seq
= new_dump_seq
;
1424 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
1426 ovs_mutex_lock(&ofproto_mutex
);
1427 LIST_FOR_EACH_SAFE (rule
, next_rule
, expirable
,
1428 &ofproto
->up
.expirable
) {
1429 rule_expire(rule_dpif_cast(rule
));
1431 ovs_mutex_unlock(&ofproto_mutex
);
1433 /* All outstanding data in existing flows has been accounted, so it's a
1434 * good time to do bond rebalancing. */
1435 if (ofproto
->has_bonded_bundles
) {
1436 struct ofbundle
*bundle
;
1438 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
1440 bond_rebalance(bundle
->bond
);
1450 wait(struct ofproto
*ofproto_
)
1452 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1454 if (ofproto_get_flow_restore_wait()) {
1458 if (ofproto
->sflow
) {
1459 dpif_sflow_wait(ofproto
->sflow
);
1461 if (ofproto
->ipfix
) {
1462 dpif_ipfix_wait(ofproto
->ipfix
);
1464 if (ofproto
->lacp_enabled
|| ofproto
->has_bonded_bundles
) {
1465 struct ofbundle
*bundle
;
1467 HMAP_FOR_EACH (bundle
, hmap_node
, &ofproto
->bundles
) {
1468 bundle_wait(bundle
);
1471 if (ofproto
->netflow
) {
1472 netflow_wait(ofproto
->netflow
);
1474 ovs_rwlock_rdlock(&ofproto
->ml
->rwlock
);
1475 mac_learning_wait(ofproto
->ml
);
1476 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
1478 if (ofproto
->backer
->need_revalidate
) {
1479 /* Shouldn't happen, but if it does just go around again. */
1480 VLOG_DBG_RL(&rl
, "need revalidate in ofproto_wait_cb()");
1481 poll_immediate_wake();
1484 seq_wait(udpif_dump_seq(ofproto
->backer
->udpif
), ofproto
->dump_seq
);
1485 seq_wait(ofproto
->pins_seq
, ofproto
->pins_seqno
);
1489 type_get_memory_usage(const char *type
, struct simap
*usage
)
1491 struct dpif_backer
*backer
;
1493 backer
= shash_find_data(&all_dpif_backers
, type
);
1495 udpif_get_memory_usage(backer
->udpif
, usage
);
1500 flush(struct ofproto
*ofproto_
)
1502 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1503 struct dpif_backer
*backer
= ofproto
->backer
;
1506 udpif_flush(backer
->udpif
);
1511 get_features(struct ofproto
*ofproto_ OVS_UNUSED
,
1512 bool *arp_match_ip
, enum ofputil_action_bitmap
*actions
)
1514 *arp_match_ip
= true;
1515 *actions
= (OFPUTIL_A_OUTPUT
|
1516 OFPUTIL_A_SET_VLAN_VID
|
1517 OFPUTIL_A_SET_VLAN_PCP
|
1518 OFPUTIL_A_STRIP_VLAN
|
1519 OFPUTIL_A_SET_DL_SRC
|
1520 OFPUTIL_A_SET_DL_DST
|
1521 OFPUTIL_A_SET_NW_SRC
|
1522 OFPUTIL_A_SET_NW_DST
|
1523 OFPUTIL_A_SET_NW_TOS
|
1524 OFPUTIL_A_SET_TP_SRC
|
1525 OFPUTIL_A_SET_TP_DST
|
1530 get_tables(struct ofproto
*ofproto_
, struct ofp12_table_stats
*ots
)
1532 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1533 struct dpif_dp_stats s
;
1534 uint64_t n_miss
, n_no_pkt_in
, n_bytes
, n_dropped_frags
;
1538 strcpy(ots
->name
, "classifier");
1540 dpif_get_dp_stats(ofproto
->backer
->dpif
, &s
);
1541 rule_get_stats(&ofproto
->miss_rule
->up
, &n_miss
, &n_bytes
, &used
);
1542 rule_get_stats(&ofproto
->no_packet_in_rule
->up
, &n_no_pkt_in
, &n_bytes
,
1544 rule_get_stats(&ofproto
->drop_frags_rule
->up
, &n_dropped_frags
, &n_bytes
,
1546 n_lookup
= s
.n_hit
+ s
.n_missed
- n_dropped_frags
;
1547 ots
->lookup_count
= htonll(n_lookup
);
1548 ots
->matched_count
= htonll(n_lookup
- n_miss
- n_no_pkt_in
);
1551 static struct ofport
*
1554 struct ofport_dpif
*port
= xmalloc(sizeof *port
);
1559 port_dealloc(struct ofport
*port_
)
1561 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
1566 port_construct(struct ofport
*port_
)
1568 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
1569 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
1570 const struct netdev
*netdev
= port
->up
.netdev
;
1571 char namebuf
[NETDEV_VPORT_NAME_BUFSIZE
];
1572 struct dpif_port dpif_port
;
1575 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1576 port
->bundle
= NULL
;
1579 port
->may_enable
= true;
1580 port
->stp_port
= NULL
;
1581 port
->stp_state
= STP_DISABLED
;
1582 port
->is_tunnel
= false;
1586 port
->realdev_ofp_port
= 0;
1587 port
->vlandev_vid
= 0;
1588 port
->carrier_seq
= netdev_get_carrier_resets(netdev
);
1589 port
->is_layer3
= netdev_vport_is_layer3(netdev
);
1591 if (netdev_vport_is_patch(netdev
)) {
1592 /* By bailing out here, we don't submit the port to the sFlow module
1593 * to be considered for counter polling export. This is correct
1594 * because the patch port represents an interface that sFlow considers
1595 * to be "internal" to the switch as a whole, and therefore not an
1596 * candidate for counter polling. */
1597 port
->odp_port
= ODPP_NONE
;
1598 ofport_update_peer(port
);
1602 error
= dpif_port_query_by_name(ofproto
->backer
->dpif
,
1603 netdev_vport_get_dpif_port(netdev
, namebuf
,
1610 port
->odp_port
= dpif_port
.port_no
;
1612 if (netdev_get_tunnel_config(netdev
)) {
1613 tnl_port_add(port
, port
->up
.netdev
, port
->odp_port
);
1614 port
->is_tunnel
= true;
1616 /* Sanity-check that a mapping doesn't already exist. This
1617 * shouldn't happen for non-tunnel ports. */
1618 if (odp_port_to_ofp_port(ofproto
, port
->odp_port
) != OFPP_NONE
) {
1619 VLOG_ERR("port %s already has an OpenFlow port number",
1621 dpif_port_destroy(&dpif_port
);
1625 ovs_rwlock_wrlock(&ofproto
->backer
->odp_to_ofport_lock
);
1626 hmap_insert(&ofproto
->backer
->odp_to_ofport_map
, &port
->odp_port_node
,
1627 hash_odp_port(port
->odp_port
));
1628 ovs_rwlock_unlock(&ofproto
->backer
->odp_to_ofport_lock
);
1630 dpif_port_destroy(&dpif_port
);
1632 if (ofproto
->sflow
) {
1633 dpif_sflow_add_port(ofproto
->sflow
, port_
, port
->odp_port
);
1640 port_destruct(struct ofport
*port_
)
1642 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
1643 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
1644 const char *devname
= netdev_get_name(port
->up
.netdev
);
1645 char namebuf
[NETDEV_VPORT_NAME_BUFSIZE
];
1646 const char *dp_port_name
;
1648 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1649 ovs_rwlock_wrlock(&xlate_rwlock
);
1650 xlate_ofport_remove(port
);
1651 ovs_rwlock_unlock(&xlate_rwlock
);
1653 dp_port_name
= netdev_vport_get_dpif_port(port
->up
.netdev
, namebuf
,
1655 if (dpif_port_exists(ofproto
->backer
->dpif
, dp_port_name
)) {
1656 /* The underlying device is still there, so delete it. This
1657 * happens when the ofproto is being destroyed, since the caller
1658 * assumes that removal of attached ports will happen as part of
1660 if (!port
->is_tunnel
) {
1661 dpif_port_del(ofproto
->backer
->dpif
, port
->odp_port
);
1666 port
->peer
->peer
= NULL
;
1670 if (port
->odp_port
!= ODPP_NONE
&& !port
->is_tunnel
) {
1671 ovs_rwlock_wrlock(&ofproto
->backer
->odp_to_ofport_lock
);
1672 hmap_remove(&ofproto
->backer
->odp_to_ofport_map
, &port
->odp_port_node
);
1673 ovs_rwlock_unlock(&ofproto
->backer
->odp_to_ofport_lock
);
1677 sset_find_and_delete(&ofproto
->ports
, devname
);
1678 sset_find_and_delete(&ofproto
->ghost_ports
, devname
);
1679 bundle_remove(port_
);
1680 set_cfm(port_
, NULL
);
1681 set_bfd(port_
, NULL
);
1682 if (port
->stp_port
) {
1683 stp_port_disable(port
->stp_port
);
1685 if (ofproto
->sflow
) {
1686 dpif_sflow_del_port(ofproto
->sflow
, port
->odp_port
);
1693 port_modified(struct ofport
*port_
)
1695 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
1697 if (port
->bundle
&& port
->bundle
->bond
) {
1698 bond_slave_set_netdev(port
->bundle
->bond
, port
, port
->up
.netdev
);
1702 cfm_set_netdev(port
->cfm
, port
->up
.netdev
);
1706 bfd_set_netdev(port
->bfd
, port
->up
.netdev
);
1709 ofproto_dpif_monitor_port_update(port
, port
->bfd
, port
->cfm
,
1710 port
->up
.pp
.hw_addr
);
1712 if (port
->is_tunnel
&& tnl_port_reconfigure(port
, port
->up
.netdev
,
1714 ofproto_dpif_cast(port
->up
.ofproto
)->backer
->need_revalidate
=
1718 ofport_update_peer(port
);
1722 port_reconfigured(struct ofport
*port_
, enum ofputil_port_config old_config
)
1724 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
1725 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
1726 enum ofputil_port_config changed
= old_config
^ port
->up
.pp
.config
;
1728 if (changed
& (OFPUTIL_PC_NO_RECV
| OFPUTIL_PC_NO_RECV_STP
|
1729 OFPUTIL_PC_NO_FWD
| OFPUTIL_PC_NO_FLOOD
|
1730 OFPUTIL_PC_NO_PACKET_IN
)) {
1731 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1733 if (changed
& OFPUTIL_PC_NO_FLOOD
&& port
->bundle
) {
1734 bundle_update(port
->bundle
);
1740 set_sflow(struct ofproto
*ofproto_
,
1741 const struct ofproto_sflow_options
*sflow_options
)
1743 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1744 struct dpif_sflow
*ds
= ofproto
->sflow
;
1746 if (sflow_options
) {
1748 struct ofport_dpif
*ofport
;
1750 ds
= ofproto
->sflow
= dpif_sflow_create();
1751 HMAP_FOR_EACH (ofport
, up
.hmap_node
, &ofproto
->up
.ports
) {
1752 dpif_sflow_add_port(ds
, &ofport
->up
, ofport
->odp_port
);
1754 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1756 dpif_sflow_set_options(ds
, sflow_options
);
1759 dpif_sflow_unref(ds
);
1760 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1761 ofproto
->sflow
= NULL
;
1769 struct ofproto
*ofproto_
,
1770 const struct ofproto_ipfix_bridge_exporter_options
*bridge_exporter_options
,
1771 const struct ofproto_ipfix_flow_exporter_options
*flow_exporters_options
,
1772 size_t n_flow_exporters_options
)
1774 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1775 struct dpif_ipfix
*di
= ofproto
->ipfix
;
1776 bool has_options
= bridge_exporter_options
|| flow_exporters_options
;
1778 if (has_options
&& !di
) {
1779 di
= ofproto
->ipfix
= dpif_ipfix_create();
1783 /* Call set_options in any case to cleanly flush the flow
1784 * caches in the last exporters that are to be destroyed. */
1785 dpif_ipfix_set_options(
1786 di
, bridge_exporter_options
, flow_exporters_options
,
1787 n_flow_exporters_options
);
1790 dpif_ipfix_unref(di
);
1791 ofproto
->ipfix
= NULL
;
1799 set_cfm(struct ofport
*ofport_
, const struct cfm_settings
*s
)
1801 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
1806 struct ofproto_dpif
*ofproto
;
1808 ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
1809 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1810 ofport
->cfm
= cfm_create(ofport
->up
.netdev
);
1813 if (cfm_configure(ofport
->cfm
, s
)) {
1820 cfm_unref(ofport
->cfm
);
1823 ofproto_dpif_monitor_port_update(ofport
, ofport
->bfd
, ofport
->cfm
,
1824 ofport
->up
.pp
.hw_addr
);
1829 get_cfm_status(const struct ofport
*ofport_
,
1830 struct ofproto_cfm_status
*status
)
1832 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
1836 if (cfm_check_status_change(ofport
->cfm
)) {
1837 status
->faults
= cfm_get_fault(ofport
->cfm
);
1838 status
->flap_count
= cfm_get_flap_count(ofport
->cfm
);
1839 status
->remote_opstate
= cfm_get_opup(ofport
->cfm
);
1840 status
->health
= cfm_get_health(ofport
->cfm
);
1841 cfm_get_remote_mpids(ofport
->cfm
, &status
->rmps
, &status
->n_rmps
);
1843 ret
= NO_STATUS_CHANGE
;
1853 set_bfd(struct ofport
*ofport_
, const struct smap
*cfg
)
1855 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport_
->ofproto
);
1856 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
1860 ofport
->bfd
= bfd_configure(old
, netdev_get_name(ofport
->up
.netdev
),
1861 cfg
, ofport
->up
.netdev
);
1862 if (ofport
->bfd
!= old
) {
1863 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1865 ofproto_dpif_monitor_port_update(ofport
, ofport
->bfd
, ofport
->cfm
,
1866 ofport
->up
.pp
.hw_addr
);
1871 get_bfd_status(struct ofport
*ofport_
, struct smap
*smap
)
1873 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
1877 if (bfd_check_status_change(ofport
->bfd
)) {
1878 bfd_get_status(ofport
->bfd
, smap
);
1880 ret
= NO_STATUS_CHANGE
;
1889 /* Spanning Tree. */
1892 send_bpdu_cb(struct ofpbuf
*pkt
, int port_num
, void *ofproto_
)
1894 struct ofproto_dpif
*ofproto
= ofproto_
;
1895 struct stp_port
*sp
= stp_get_port(ofproto
->stp
, port_num
);
1896 struct ofport_dpif
*ofport
;
1898 ofport
= stp_port_get_aux(sp
);
1900 VLOG_WARN_RL(&rl
, "%s: cannot send BPDU on unknown port %d",
1901 ofproto
->up
.name
, port_num
);
1903 struct eth_header
*eth
= ofpbuf_l2(pkt
);
1905 netdev_get_etheraddr(ofport
->up
.netdev
, eth
->eth_src
);
1906 if (eth_addr_is_zero(eth
->eth_src
)) {
1907 VLOG_WARN_RL(&rl
, "%s: cannot send BPDU on port %d "
1908 "with unknown MAC", ofproto
->up
.name
, port_num
);
1910 ofproto_dpif_send_packet(ofport
, pkt
);
1916 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1918 set_stp(struct ofproto
*ofproto_
, const struct ofproto_stp_settings
*s
)
1920 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1922 /* Only revalidate flows if the configuration changed. */
1923 if (!s
!= !ofproto
->stp
) {
1924 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
1928 if (!ofproto
->stp
) {
1929 ofproto
->stp
= stp_create(ofproto_
->name
, s
->system_id
,
1930 send_bpdu_cb
, ofproto
);
1931 ofproto
->stp_last_tick
= time_msec();
1934 stp_set_bridge_id(ofproto
->stp
, s
->system_id
);
1935 stp_set_bridge_priority(ofproto
->stp
, s
->priority
);
1936 stp_set_hello_time(ofproto
->stp
, s
->hello_time
);
1937 stp_set_max_age(ofproto
->stp
, s
->max_age
);
1938 stp_set_forward_delay(ofproto
->stp
, s
->fwd_delay
);
1940 struct ofport
*ofport
;
1942 HMAP_FOR_EACH (ofport
, hmap_node
, &ofproto
->up
.ports
) {
1943 set_stp_port(ofport
, NULL
);
1946 stp_unref(ofproto
->stp
);
1947 ofproto
->stp
= NULL
;
1954 get_stp_status(struct ofproto
*ofproto_
, struct ofproto_stp_status
*s
)
1956 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
1960 s
->bridge_id
= stp_get_bridge_id(ofproto
->stp
);
1961 s
->designated_root
= stp_get_designated_root(ofproto
->stp
);
1962 s
->root_path_cost
= stp_get_root_path_cost(ofproto
->stp
);
1971 update_stp_port_state(struct ofport_dpif
*ofport
)
1973 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
1974 enum stp_state state
;
1976 /* Figure out new state. */
1977 state
= ofport
->stp_port
? stp_port_get_state(ofport
->stp_port
)
1981 if (ofport
->stp_state
!= state
) {
1982 enum ofputil_port_state of_state
;
1985 VLOG_DBG_RL(&rl
, "port %s: STP state changed from %s to %s",
1986 netdev_get_name(ofport
->up
.netdev
),
1987 stp_state_name(ofport
->stp_state
),
1988 stp_state_name(state
));
1989 if (stp_learn_in_state(ofport
->stp_state
)
1990 != stp_learn_in_state(state
)) {
1991 /* xxx Learning action flows should also be flushed. */
1992 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
1993 mac_learning_flush(ofproto
->ml
);
1994 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
1996 fwd_change
= stp_forward_in_state(ofport
->stp_state
)
1997 != stp_forward_in_state(state
);
1999 ofproto
->backer
->need_revalidate
= REV_STP
;
2000 ofport
->stp_state
= state
;
2001 ofport
->stp_state_entered
= time_msec();
2003 if (fwd_change
&& ofport
->bundle
) {
2004 bundle_update(ofport
->bundle
);
2007 /* Update the STP state bits in the OpenFlow port description. */
2008 of_state
= ofport
->up
.pp
.state
& ~OFPUTIL_PS_STP_MASK
;
2009 of_state
|= (state
== STP_LISTENING
? OFPUTIL_PS_STP_LISTEN
2010 : state
== STP_LEARNING
? OFPUTIL_PS_STP_LEARN
2011 : state
== STP_FORWARDING
? OFPUTIL_PS_STP_FORWARD
2012 : state
== STP_BLOCKING
? OFPUTIL_PS_STP_BLOCK
2014 ofproto_port_set_state(&ofport
->up
, of_state
);
2018 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2019 * caller is responsible for assigning STP port numbers and ensuring
2020 * there are no duplicates. */
2022 set_stp_port(struct ofport
*ofport_
,
2023 const struct ofproto_port_stp_settings
*s
)
2025 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2026 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2027 struct stp_port
*sp
= ofport
->stp_port
;
2029 if (!s
|| !s
->enable
) {
2031 ofport
->stp_port
= NULL
;
2032 stp_port_disable(sp
);
2033 update_stp_port_state(ofport
);
2036 } else if (sp
&& stp_port_no(sp
) != s
->port_num
2037 && ofport
== stp_port_get_aux(sp
)) {
2038 /* The port-id changed, so disable the old one if it's not
2039 * already in use by another port. */
2040 stp_port_disable(sp
);
2043 sp
= ofport
->stp_port
= stp_get_port(ofproto
->stp
, s
->port_num
);
2044 stp_port_enable(sp
);
2046 stp_port_set_aux(sp
, ofport
);
2047 stp_port_set_priority(sp
, s
->priority
);
2048 stp_port_set_path_cost(sp
, s
->path_cost
);
2050 update_stp_port_state(ofport
);
2056 get_stp_port_status(struct ofport
*ofport_
,
2057 struct ofproto_port_stp_status
*s
)
2059 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2060 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2061 struct stp_port
*sp
= ofport
->stp_port
;
2063 if (!ofproto
->stp
|| !sp
) {
2069 s
->port_id
= stp_port_get_id(sp
);
2070 s
->state
= stp_port_get_state(sp
);
2071 s
->sec_in_state
= (time_msec() - ofport
->stp_state_entered
) / 1000;
2072 s
->role
= stp_port_get_role(sp
);
2078 get_stp_port_stats(struct ofport
*ofport_
,
2079 struct ofproto_port_stp_stats
*s
)
2081 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2082 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2083 struct stp_port
*sp
= ofport
->stp_port
;
2085 if (!ofproto
->stp
|| !sp
) {
2091 stp_port_get_counts(sp
, &s
->tx_count
, &s
->rx_count
, &s
->error_count
);
2097 stp_run(struct ofproto_dpif
*ofproto
)
2100 long long int now
= time_msec();
2101 long long int elapsed
= now
- ofproto
->stp_last_tick
;
2102 struct stp_port
*sp
;
2105 stp_tick(ofproto
->stp
, MIN(INT_MAX
, elapsed
));
2106 ofproto
->stp_last_tick
= now
;
2108 while (stp_get_changed_port(ofproto
->stp
, &sp
)) {
2109 struct ofport_dpif
*ofport
= stp_port_get_aux(sp
);
2112 update_stp_port_state(ofport
);
2116 if (stp_check_and_reset_fdb_flush(ofproto
->stp
)) {
2117 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
2118 mac_learning_flush(ofproto
->ml
);
2119 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
2125 stp_wait(struct ofproto_dpif
*ofproto
)
2128 poll_timer_wait(1000);
2133 set_queues(struct ofport
*ofport_
, const struct ofproto_port_queue
*qdscp
,
2136 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2137 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2139 if (ofport
->n_qdscp
!= n_qdscp
2140 || (n_qdscp
&& memcmp(ofport
->qdscp
, qdscp
,
2141 n_qdscp
* sizeof *qdscp
))) {
2142 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2143 free(ofport
->qdscp
);
2144 ofport
->qdscp
= n_qdscp
2145 ? xmemdup(qdscp
, n_qdscp
* sizeof *qdscp
)
2147 ofport
->n_qdscp
= n_qdscp
;
2155 /* Expires all MAC learning entries associated with 'bundle' and forces its
2156 * ofproto to revalidate every flow.
2158 * Normally MAC learning entries are removed only from the ofproto associated
2159 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2160 * are removed from every ofproto. When patch ports and SLB bonds are in use
2161 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2162 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2163 * with the host from which it migrated. */
2165 bundle_flush_macs(struct ofbundle
*bundle
, bool all_ofprotos
)
2167 struct ofproto_dpif
*ofproto
= bundle
->ofproto
;
2168 struct mac_learning
*ml
= ofproto
->ml
;
2169 struct mac_entry
*mac
, *next_mac
;
2171 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2172 ovs_rwlock_wrlock(&ml
->rwlock
);
2173 LIST_FOR_EACH_SAFE (mac
, next_mac
, lru_node
, &ml
->lrus
) {
2174 if (mac
->port
.p
== bundle
) {
2176 struct ofproto_dpif
*o
;
2178 HMAP_FOR_EACH (o
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
2180 struct mac_entry
*e
;
2182 ovs_rwlock_wrlock(&o
->ml
->rwlock
);
2183 e
= mac_learning_lookup(o
->ml
, mac
->mac
, mac
->vlan
);
2185 mac_learning_expire(o
->ml
, e
);
2187 ovs_rwlock_unlock(&o
->ml
->rwlock
);
2192 mac_learning_expire(ml
, mac
);
2195 ovs_rwlock_unlock(&ml
->rwlock
);
2198 static struct ofbundle
*
2199 bundle_lookup(const struct ofproto_dpif
*ofproto
, void *aux
)
2201 struct ofbundle
*bundle
;
2203 HMAP_FOR_EACH_IN_BUCKET (bundle
, hmap_node
, hash_pointer(aux
, 0),
2204 &ofproto
->bundles
) {
2205 if (bundle
->aux
== aux
) {
2213 bundle_update(struct ofbundle
*bundle
)
2215 struct ofport_dpif
*port
;
2217 bundle
->floodable
= true;
2218 LIST_FOR_EACH (port
, bundle_node
, &bundle
->ports
) {
2219 if (port
->up
.pp
.config
& OFPUTIL_PC_NO_FLOOD
2221 || !stp_forward_in_state(port
->stp_state
)) {
2222 bundle
->floodable
= false;
2229 bundle_del_port(struct ofport_dpif
*port
)
2231 struct ofbundle
*bundle
= port
->bundle
;
2233 bundle
->ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2235 list_remove(&port
->bundle_node
);
2236 port
->bundle
= NULL
;
2239 lacp_slave_unregister(bundle
->lacp
, port
);
2242 bond_slave_unregister(bundle
->bond
, port
);
2245 bundle_update(bundle
);
2249 bundle_add_port(struct ofbundle
*bundle
, ofp_port_t ofp_port
,
2250 struct lacp_slave_settings
*lacp
)
2252 struct ofport_dpif
*port
;
2254 port
= get_ofp_port(bundle
->ofproto
, ofp_port
);
2259 if (port
->bundle
!= bundle
) {
2260 bundle
->ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2262 bundle_remove(&port
->up
);
2265 port
->bundle
= bundle
;
2266 list_push_back(&bundle
->ports
, &port
->bundle_node
);
2267 if (port
->up
.pp
.config
& OFPUTIL_PC_NO_FLOOD
2269 || !stp_forward_in_state(port
->stp_state
)) {
2270 bundle
->floodable
= false;
2274 bundle
->ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2275 lacp_slave_register(bundle
->lacp
, port
, lacp
);
2282 bundle_destroy(struct ofbundle
*bundle
)
2284 struct ofproto_dpif
*ofproto
;
2285 struct ofport_dpif
*port
, *next_port
;
2291 ofproto
= bundle
->ofproto
;
2292 mbridge_unregister_bundle(ofproto
->mbridge
, bundle
->aux
);
2294 ovs_rwlock_wrlock(&xlate_rwlock
);
2295 xlate_bundle_remove(bundle
);
2296 ovs_rwlock_unlock(&xlate_rwlock
);
2298 LIST_FOR_EACH_SAFE (port
, next_port
, bundle_node
, &bundle
->ports
) {
2299 bundle_del_port(port
);
2302 bundle_flush_macs(bundle
, true);
2303 hmap_remove(&ofproto
->bundles
, &bundle
->hmap_node
);
2305 free(bundle
->trunks
);
2306 lacp_unref(bundle
->lacp
);
2307 bond_unref(bundle
->bond
);
2312 bundle_set(struct ofproto
*ofproto_
, void *aux
,
2313 const struct ofproto_bundle_settings
*s
)
2315 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2316 bool need_flush
= false;
2317 struct ofport_dpif
*port
;
2318 struct ofbundle
*bundle
;
2319 unsigned long *trunks
;
2325 bundle_destroy(bundle_lookup(ofproto
, aux
));
2329 ovs_assert(s
->n_slaves
== 1 || s
->bond
!= NULL
);
2330 ovs_assert((s
->lacp
!= NULL
) == (s
->lacp_slaves
!= NULL
));
2332 bundle
= bundle_lookup(ofproto
, aux
);
2334 bundle
= xmalloc(sizeof *bundle
);
2336 bundle
->ofproto
= ofproto
;
2337 hmap_insert(&ofproto
->bundles
, &bundle
->hmap_node
,
2338 hash_pointer(aux
, 0));
2340 bundle
->name
= NULL
;
2342 list_init(&bundle
->ports
);
2343 bundle
->vlan_mode
= PORT_VLAN_TRUNK
;
2345 bundle
->trunks
= NULL
;
2346 bundle
->use_priority_tags
= s
->use_priority_tags
;
2347 bundle
->lacp
= NULL
;
2348 bundle
->bond
= NULL
;
2350 bundle
->floodable
= true;
2351 mbridge_register_bundle(ofproto
->mbridge
, bundle
);
2354 if (!bundle
->name
|| strcmp(s
->name
, bundle
->name
)) {
2356 bundle
->name
= xstrdup(s
->name
);
2361 ofproto
->lacp_enabled
= true;
2362 if (!bundle
->lacp
) {
2363 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2364 bundle
->lacp
= lacp_create();
2366 lacp_configure(bundle
->lacp
, s
->lacp
);
2368 lacp_unref(bundle
->lacp
);
2369 bundle
->lacp
= NULL
;
2372 /* Update set of ports. */
2374 for (i
= 0; i
< s
->n_slaves
; i
++) {
2375 if (!bundle_add_port(bundle
, s
->slaves
[i
],
2376 s
->lacp
? &s
->lacp_slaves
[i
] : NULL
)) {
2380 if (!ok
|| list_size(&bundle
->ports
) != s
->n_slaves
) {
2381 struct ofport_dpif
*next_port
;
2383 LIST_FOR_EACH_SAFE (port
, next_port
, bundle_node
, &bundle
->ports
) {
2384 for (i
= 0; i
< s
->n_slaves
; i
++) {
2385 if (s
->slaves
[i
] == port
->up
.ofp_port
) {
2390 bundle_del_port(port
);
2394 ovs_assert(list_size(&bundle
->ports
) <= s
->n_slaves
);
2396 if (list_is_empty(&bundle
->ports
)) {
2397 bundle_destroy(bundle
);
2401 /* Set VLAN tagging mode */
2402 if (s
->vlan_mode
!= bundle
->vlan_mode
2403 || s
->use_priority_tags
!= bundle
->use_priority_tags
) {
2404 bundle
->vlan_mode
= s
->vlan_mode
;
2405 bundle
->use_priority_tags
= s
->use_priority_tags
;
2410 vlan
= (s
->vlan_mode
== PORT_VLAN_TRUNK
? -1
2411 : s
->vlan
>= 0 && s
->vlan
<= 4095 ? s
->vlan
2413 if (vlan
!= bundle
->vlan
) {
2414 bundle
->vlan
= vlan
;
2418 /* Get trunked VLANs. */
2419 switch (s
->vlan_mode
) {
2420 case PORT_VLAN_ACCESS
:
2424 case PORT_VLAN_TRUNK
:
2425 trunks
= CONST_CAST(unsigned long *, s
->trunks
);
2428 case PORT_VLAN_NATIVE_UNTAGGED
:
2429 case PORT_VLAN_NATIVE_TAGGED
:
2430 if (vlan
!= 0 && (!s
->trunks
2431 || !bitmap_is_set(s
->trunks
, vlan
)
2432 || bitmap_is_set(s
->trunks
, 0))) {
2433 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2435 trunks
= bitmap_clone(s
->trunks
, 4096);
2437 trunks
= bitmap_allocate1(4096);
2439 bitmap_set1(trunks
, vlan
);
2440 bitmap_set0(trunks
, 0);
2442 trunks
= CONST_CAST(unsigned long *, s
->trunks
);
2449 if (!vlan_bitmap_equal(trunks
, bundle
->trunks
)) {
2450 free(bundle
->trunks
);
2451 if (trunks
== s
->trunks
) {
2452 bundle
->trunks
= vlan_bitmap_clone(trunks
);
2454 bundle
->trunks
= trunks
;
2459 if (trunks
!= s
->trunks
) {
2464 if (!list_is_short(&bundle
->ports
)) {
2465 bundle
->ofproto
->has_bonded_bundles
= true;
2467 if (bond_reconfigure(bundle
->bond
, s
->bond
)) {
2468 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2471 bundle
->bond
= bond_create(s
->bond
, ofproto
);
2472 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2475 LIST_FOR_EACH (port
, bundle_node
, &bundle
->ports
) {
2476 bond_slave_register(bundle
->bond
, port
,
2477 port
->up
.ofp_port
, port
->up
.netdev
);
2480 bond_unref(bundle
->bond
);
2481 bundle
->bond
= NULL
;
2484 /* If we changed something that would affect MAC learning, un-learn
2485 * everything on this port and force flow revalidation. */
2487 bundle_flush_macs(bundle
, false);
2494 bundle_remove(struct ofport
*port_
)
2496 struct ofport_dpif
*port
= ofport_dpif_cast(port_
);
2497 struct ofbundle
*bundle
= port
->bundle
;
2500 bundle_del_port(port
);
2501 if (list_is_empty(&bundle
->ports
)) {
2502 bundle_destroy(bundle
);
2503 } else if (list_is_short(&bundle
->ports
)) {
2504 bond_unref(bundle
->bond
);
2505 bundle
->bond
= NULL
;
2511 send_pdu_cb(void *port_
, const void *pdu
, size_t pdu_size
)
2513 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 10);
2514 struct ofport_dpif
*port
= port_
;
2515 uint8_t ea
[ETH_ADDR_LEN
];
2518 error
= netdev_get_etheraddr(port
->up
.netdev
, ea
);
2520 struct ofpbuf packet
;
2523 ofpbuf_init(&packet
, 0);
2524 packet_pdu
= eth_compose(&packet
, eth_addr_lacp
, ea
, ETH_TYPE_LACP
,
2526 memcpy(packet_pdu
, pdu
, pdu_size
);
2528 ofproto_dpif_send_packet(port
, &packet
);
2529 ofpbuf_uninit(&packet
);
2531 VLOG_ERR_RL(&rl
, "port %s: cannot obtain Ethernet address of iface "
2532 "%s (%s)", port
->bundle
->name
,
2533 netdev_get_name(port
->up
.netdev
), ovs_strerror(error
));
2538 bundle_send_learning_packets(struct ofbundle
*bundle
)
2540 struct ofproto_dpif
*ofproto
= bundle
->ofproto
;
2541 struct ofpbuf
*learning_packet
;
2542 int error
, n_packets
, n_errors
;
2543 struct mac_entry
*e
;
2544 struct list packets
;
2546 list_init(&packets
);
2547 ovs_rwlock_rdlock(&ofproto
->ml
->rwlock
);
2548 LIST_FOR_EACH (e
, lru_node
, &ofproto
->ml
->lrus
) {
2549 if (e
->port
.p
!= bundle
) {
2552 learning_packet
= bond_compose_learning_packet(bundle
->bond
,
2555 /* Temporarily use 'frame' as a private pointer (see below). */
2556 ovs_assert(learning_packet
->frame
== ofpbuf_data(learning_packet
));
2557 learning_packet
->frame
= port_void
;
2558 list_push_back(&packets
, &learning_packet
->list_node
);
2561 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
2563 error
= n_packets
= n_errors
= 0;
2564 LIST_FOR_EACH (learning_packet
, list_node
, &packets
) {
2566 void *port_void
= learning_packet
->frame
;
2568 /* Restore 'frame'. */
2569 learning_packet
->frame
= ofpbuf_data(learning_packet
);
2570 ret
= ofproto_dpif_send_packet(port_void
, learning_packet
);
2577 ofpbuf_list_delete(&packets
);
2580 static struct vlog_rate_limit rl
= VLOG_RATE_LIMIT_INIT(1, 5);
2581 VLOG_WARN_RL(&rl
, "bond %s: %d errors sending %d gratuitous learning "
2582 "packets, last error was: %s",
2583 bundle
->name
, n_errors
, n_packets
, ovs_strerror(error
));
2585 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2586 bundle
->name
, n_packets
);
2591 bundle_run(struct ofbundle
*bundle
)
2594 lacp_run(bundle
->lacp
, send_pdu_cb
);
2597 struct ofport_dpif
*port
;
2599 LIST_FOR_EACH (port
, bundle_node
, &bundle
->ports
) {
2600 bond_slave_set_may_enable(bundle
->bond
, port
, port
->may_enable
);
2603 if (bond_run(bundle
->bond
, lacp_status(bundle
->lacp
))) {
2604 bundle
->ofproto
->backer
->need_revalidate
= REV_BOND
;
2607 if (bond_should_send_learning_packets(bundle
->bond
)) {
2608 bundle_send_learning_packets(bundle
);
2614 bundle_wait(struct ofbundle
*bundle
)
2617 lacp_wait(bundle
->lacp
);
2620 bond_wait(bundle
->bond
);
2627 mirror_set__(struct ofproto
*ofproto_
, void *aux
,
2628 const struct ofproto_mirror_settings
*s
)
2630 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2631 struct ofbundle
**srcs
, **dsts
;
2636 mirror_destroy(ofproto
->mbridge
, aux
);
2640 srcs
= xmalloc(s
->n_srcs
* sizeof *srcs
);
2641 dsts
= xmalloc(s
->n_dsts
* sizeof *dsts
);
2643 for (i
= 0; i
< s
->n_srcs
; i
++) {
2644 srcs
[i
] = bundle_lookup(ofproto
, s
->srcs
[i
]);
2647 for (i
= 0; i
< s
->n_dsts
; i
++) {
2648 dsts
[i
] = bundle_lookup(ofproto
, s
->dsts
[i
]);
2651 error
= mirror_set(ofproto
->mbridge
, aux
, s
->name
, srcs
, s
->n_srcs
, dsts
,
2652 s
->n_dsts
, s
->src_vlans
,
2653 bundle_lookup(ofproto
, s
->out_bundle
), s
->out_vlan
);
2660 mirror_get_stats__(struct ofproto
*ofproto
, void *aux
,
2661 uint64_t *packets
, uint64_t *bytes
)
2663 return mirror_get_stats(ofproto_dpif_cast(ofproto
)->mbridge
, aux
, packets
,
2668 set_flood_vlans(struct ofproto
*ofproto_
, unsigned long *flood_vlans
)
2670 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2671 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
2672 if (mac_learning_set_flood_vlans(ofproto
->ml
, flood_vlans
)) {
2673 mac_learning_flush(ofproto
->ml
);
2675 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
2680 is_mirror_output_bundle(const struct ofproto
*ofproto_
, void *aux
)
2682 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2683 struct ofbundle
*bundle
= bundle_lookup(ofproto
, aux
);
2684 return bundle
&& mirror_bundle_out(ofproto
->mbridge
, bundle
) != 0;
2688 forward_bpdu_changed(struct ofproto
*ofproto_
)
2690 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2691 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2695 set_mac_table_config(struct ofproto
*ofproto_
, unsigned int idle_time
,
2698 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2699 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
2700 mac_learning_set_idle_time(ofproto
->ml
, idle_time
);
2701 mac_learning_set_max_entries(ofproto
->ml
, max_entries
);
2702 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
2707 static struct ofport_dpif
*
2708 get_ofp_port(const struct ofproto_dpif
*ofproto
, ofp_port_t ofp_port
)
2710 struct ofport
*ofport
= ofproto_get_port(&ofproto
->up
, ofp_port
);
2711 return ofport
? ofport_dpif_cast(ofport
) : NULL
;
2715 ofproto_port_from_dpif_port(struct ofproto_dpif
*ofproto
,
2716 struct ofproto_port
*ofproto_port
,
2717 struct dpif_port
*dpif_port
)
2719 ofproto_port
->name
= dpif_port
->name
;
2720 ofproto_port
->type
= dpif_port
->type
;
2721 ofproto_port
->ofp_port
= odp_port_to_ofp_port(ofproto
, dpif_port
->port_no
);
2725 ofport_update_peer(struct ofport_dpif
*ofport
)
2727 const struct ofproto_dpif
*ofproto
;
2728 struct dpif_backer
*backer
;
2731 if (!netdev_vport_is_patch(ofport
->up
.netdev
)) {
2735 backer
= ofproto_dpif_cast(ofport
->up
.ofproto
)->backer
;
2736 backer
->need_revalidate
= REV_RECONFIGURE
;
2739 ofport
->peer
->peer
= NULL
;
2740 ofport
->peer
= NULL
;
2743 peer_name
= netdev_vport_patch_peer(ofport
->up
.netdev
);
2748 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
2749 struct ofport
*peer_ofport
;
2750 struct ofport_dpif
*peer
;
2753 if (ofproto
->backer
!= backer
) {
2757 peer_ofport
= shash_find_data(&ofproto
->up
.port_by_name
, peer_name
);
2762 peer
= ofport_dpif_cast(peer_ofport
);
2763 peer_peer
= netdev_vport_patch_peer(peer
->up
.netdev
);
2764 if (peer_peer
&& !strcmp(netdev_get_name(ofport
->up
.netdev
),
2766 ofport
->peer
= peer
;
2767 ofport
->peer
->peer
= ofport
;
2777 port_run(struct ofport_dpif
*ofport
)
2779 long long int carrier_seq
= netdev_get_carrier_resets(ofport
->up
.netdev
);
2780 bool carrier_changed
= carrier_seq
!= ofport
->carrier_seq
;
2781 bool enable
= netdev_get_carrier(ofport
->up
.netdev
);
2782 bool cfm_enable
= false;
2783 bool bfd_enable
= false;
2785 ofport
->carrier_seq
= carrier_seq
;
2788 int cfm_opup
= cfm_get_opup(ofport
->cfm
);
2790 cfm_enable
= !cfm_get_fault(ofport
->cfm
);
2792 if (cfm_opup
>= 0) {
2793 cfm_enable
= cfm_enable
&& cfm_opup
;
2798 bfd_enable
= bfd_forwarding(ofport
->bfd
);
2801 if (ofport
->bfd
|| ofport
->cfm
) {
2802 enable
= enable
&& (cfm_enable
|| bfd_enable
);
2805 if (ofport
->bundle
) {
2806 enable
= enable
&& lacp_slave_may_enable(ofport
->bundle
->lacp
, ofport
);
2807 if (carrier_changed
) {
2808 lacp_slave_carrier_changed(ofport
->bundle
->lacp
, ofport
);
2812 if (ofport
->may_enable
!= enable
) {
2813 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2814 ofproto
->backer
->need_revalidate
= REV_PORT_TOGGLED
;
2817 ofport
->may_enable
= enable
;
2821 port_query_by_name(const struct ofproto
*ofproto_
, const char *devname
,
2822 struct ofproto_port
*ofproto_port
)
2824 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2825 struct dpif_port dpif_port
;
2828 if (sset_contains(&ofproto
->ghost_ports
, devname
)) {
2829 const char *type
= netdev_get_type_from_name(devname
);
2831 /* We may be called before ofproto->up.port_by_name is populated with
2832 * the appropriate ofport. For this reason, we must get the name and
2833 * type from the netdev layer directly. */
2835 const struct ofport
*ofport
;
2837 ofport
= shash_find_data(&ofproto
->up
.port_by_name
, devname
);
2838 ofproto_port
->ofp_port
= ofport
? ofport
->ofp_port
: OFPP_NONE
;
2839 ofproto_port
->name
= xstrdup(devname
);
2840 ofproto_port
->type
= xstrdup(type
);
2846 if (!sset_contains(&ofproto
->ports
, devname
)) {
2849 error
= dpif_port_query_by_name(ofproto
->backer
->dpif
,
2850 devname
, &dpif_port
);
2852 ofproto_port_from_dpif_port(ofproto
, ofproto_port
, &dpif_port
);
2858 port_add(struct ofproto
*ofproto_
, struct netdev
*netdev
)
2860 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2861 const char *devname
= netdev_get_name(netdev
);
2862 char namebuf
[NETDEV_VPORT_NAME_BUFSIZE
];
2863 const char *dp_port_name
;
2865 if (netdev_vport_is_patch(netdev
)) {
2866 sset_add(&ofproto
->ghost_ports
, netdev_get_name(netdev
));
2870 dp_port_name
= netdev_vport_get_dpif_port(netdev
, namebuf
, sizeof namebuf
);
2871 if (!dpif_port_exists(ofproto
->backer
->dpif
, dp_port_name
)) {
2872 odp_port_t port_no
= ODPP_NONE
;
2875 error
= dpif_port_add(ofproto
->backer
->dpif
, netdev
, &port_no
);
2879 if (netdev_get_tunnel_config(netdev
)) {
2880 simap_put(&ofproto
->backer
->tnl_backers
,
2881 dp_port_name
, odp_to_u32(port_no
));
2885 if (netdev_get_tunnel_config(netdev
)) {
2886 sset_add(&ofproto
->ghost_ports
, devname
);
2888 sset_add(&ofproto
->ports
, devname
);
2894 port_del(struct ofproto
*ofproto_
, ofp_port_t ofp_port
)
2896 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2897 struct ofport_dpif
*ofport
= get_ofp_port(ofproto
, ofp_port
);
2904 sset_find_and_delete(&ofproto
->ghost_ports
,
2905 netdev_get_name(ofport
->up
.netdev
));
2906 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
2907 if (!ofport
->is_tunnel
&& !netdev_vport_is_patch(ofport
->up
.netdev
)) {
2908 error
= dpif_port_del(ofproto
->backer
->dpif
, ofport
->odp_port
);
2910 /* The caller is going to close ofport->up.netdev. If this is a
2911 * bonded port, then the bond is using that netdev, so remove it
2912 * from the bond. The client will need to reconfigure everything
2913 * after deleting ports, so then the slave will get re-added. */
2914 bundle_remove(&ofport
->up
);
2921 port_get_stats(const struct ofport
*ofport_
, struct netdev_stats
*stats
)
2923 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
2926 error
= netdev_get_stats(ofport
->up
.netdev
, stats
);
2928 if (!error
&& ofport_
->ofp_port
== OFPP_LOCAL
) {
2929 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
2931 ovs_mutex_lock(&ofproto
->stats_mutex
);
2932 /* ofproto->stats.tx_packets represents packets that we created
2933 * internally and sent to some port (e.g. packets sent with
2934 * ofproto_dpif_send_packet()). Account for them as if they had
2935 * come from OFPP_LOCAL and got forwarded. */
2937 if (stats
->rx_packets
!= UINT64_MAX
) {
2938 stats
->rx_packets
+= ofproto
->stats
.tx_packets
;
2941 if (stats
->rx_bytes
!= UINT64_MAX
) {
2942 stats
->rx_bytes
+= ofproto
->stats
.tx_bytes
;
2945 /* ofproto->stats.rx_packets represents packets that were received on
2946 * some port and we processed internally and dropped (e.g. STP).
2947 * Account for them as if they had been forwarded to OFPP_LOCAL. */
2949 if (stats
->tx_packets
!= UINT64_MAX
) {
2950 stats
->tx_packets
+= ofproto
->stats
.rx_packets
;
2953 if (stats
->tx_bytes
!= UINT64_MAX
) {
2954 stats
->tx_bytes
+= ofproto
->stats
.rx_bytes
;
2956 ovs_mutex_unlock(&ofproto
->stats_mutex
);
2962 struct port_dump_state
{
2967 struct ofproto_port port
;
2972 port_dump_start(const struct ofproto
*ofproto_ OVS_UNUSED
, void **statep
)
2974 *statep
= xzalloc(sizeof(struct port_dump_state
));
2979 port_dump_next(const struct ofproto
*ofproto_
, void *state_
,
2980 struct ofproto_port
*port
)
2982 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
2983 struct port_dump_state
*state
= state_
;
2984 const struct sset
*sset
;
2985 struct sset_node
*node
;
2987 if (state
->has_port
) {
2988 ofproto_port_destroy(&state
->port
);
2989 state
->has_port
= false;
2991 sset
= state
->ghost
? &ofproto
->ghost_ports
: &ofproto
->ports
;
2992 while ((node
= sset_at_position(sset
, &state
->bucket
, &state
->offset
))) {
2995 error
= port_query_by_name(ofproto_
, node
->name
, &state
->port
);
2997 *port
= state
->port
;
2998 state
->has_port
= true;
3000 } else if (error
!= ENODEV
) {
3005 if (!state
->ghost
) {
3006 state
->ghost
= true;
3009 return port_dump_next(ofproto_
, state_
, port
);
3016 port_dump_done(const struct ofproto
*ofproto_ OVS_UNUSED
, void *state_
)
3018 struct port_dump_state
*state
= state_
;
3020 if (state
->has_port
) {
3021 ofproto_port_destroy(&state
->port
);
3028 port_poll(const struct ofproto
*ofproto_
, char **devnamep
)
3030 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3032 if (ofproto
->port_poll_errno
) {
3033 int error
= ofproto
->port_poll_errno
;
3034 ofproto
->port_poll_errno
= 0;
3038 if (sset_is_empty(&ofproto
->port_poll_set
)) {
3042 *devnamep
= sset_pop(&ofproto
->port_poll_set
);
3047 port_poll_wait(const struct ofproto
*ofproto_
)
3049 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3050 dpif_port_poll_wait(ofproto
->backer
->dpif
);
3054 port_is_lacp_current(const struct ofport
*ofport_
)
3056 const struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
3057 return (ofport
->bundle
&& ofport
->bundle
->lacp
3058 ? lacp_slave_is_current(ofport
->bundle
->lacp
, ofport
)
3062 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
3063 * then delete it entirely. */
3065 rule_expire(struct rule_dpif
*rule
)
3066 OVS_REQUIRES(ofproto_mutex
)
3068 uint16_t hard_timeout
, idle_timeout
;
3069 long long int now
= time_msec();
3072 ovs_assert(!rule
->up
.pending
);
3074 hard_timeout
= rule
->up
.hard_timeout
;
3075 idle_timeout
= rule
->up
.idle_timeout
;
3077 /* Has 'rule' expired? */
3079 long long int modified
;
3081 ovs_mutex_lock(&rule
->up
.mutex
);
3082 modified
= rule
->up
.modified
;
3083 ovs_mutex_unlock(&rule
->up
.mutex
);
3085 if (now
> modified
+ hard_timeout
* 1000) {
3086 reason
= OFPRR_HARD_TIMEOUT
;
3090 if (reason
< 0 && idle_timeout
) {
3093 ovs_mutex_lock(&rule
->stats_mutex
);
3094 used
= rule
->stats
.used
;
3095 ovs_mutex_unlock(&rule
->stats_mutex
);
3097 if (now
> used
+ idle_timeout
* 1000) {
3098 reason
= OFPRR_IDLE_TIMEOUT
;
3103 COVERAGE_INC(ofproto_dpif_expired
);
3104 ofproto_rule_expire(&rule
->up
, reason
);
3108 /* Executes, within 'ofproto', the actions in 'rule' or 'ofpacts' on 'packet'.
3109 * 'flow' must reflect the data in 'packet'. */
3111 ofproto_dpif_execute_actions(struct ofproto_dpif
*ofproto
,
3112 const struct flow
*flow
,
3113 struct rule_dpif
*rule
,
3114 const struct ofpact
*ofpacts
, size_t ofpacts_len
,
3115 struct ofpbuf
*packet
)
3117 struct dpif_flow_stats stats
;
3118 struct xlate_out xout
;
3119 struct xlate_in xin
;
3121 struct dpif_execute execute
;
3124 ovs_assert((rule
!= NULL
) != (ofpacts
!= NULL
));
3126 dpif_flow_stats_extract(flow
, packet
, time_msec(), &stats
);
3129 rule_dpif_credit_stats(rule
, &stats
);
3132 xlate_in_init(&xin
, ofproto
, flow
, rule
, stats
.tcp_flags
, packet
);
3133 xin
.ofpacts
= ofpacts
;
3134 xin
.ofpacts_len
= ofpacts_len
;
3135 xin
.resubmit_stats
= &stats
;
3136 xlate_actions(&xin
, &xout
);
3138 in_port
= flow
->in_port
.ofp_port
;
3139 if (in_port
== OFPP_NONE
) {
3140 in_port
= OFPP_LOCAL
;
3142 execute
.actions
= ofpbuf_data(&xout
.odp_actions
);
3143 execute
.actions_len
= ofpbuf_size(&xout
.odp_actions
);
3144 execute
.packet
= packet
;
3145 execute
.md
.tunnel
= flow
->tunnel
;
3146 execute
.md
.skb_priority
= flow
->skb_priority
;
3147 execute
.md
.pkt_mark
= flow
->pkt_mark
;
3148 execute
.md
.in_port
.odp_port
= ofp_port_to_odp_port(ofproto
, in_port
);
3149 execute
.needs_help
= (xout
.slow
& SLOW_ACTION
) != 0;
3151 error
= dpif_execute(ofproto
->backer
->dpif
, &execute
);
3153 xlate_out_uninit(&xout
);
3159 rule_dpif_credit_stats(struct rule_dpif
*rule
,
3160 const struct dpif_flow_stats
*stats
)
3162 ovs_mutex_lock(&rule
->stats_mutex
);
3163 rule
->stats
.n_packets
+= stats
->n_packets
;
3164 rule
->stats
.n_bytes
+= stats
->n_bytes
;
3165 rule
->stats
.used
= MAX(rule
->stats
.used
, stats
->used
);
3166 ovs_mutex_unlock(&rule
->stats_mutex
);
3170 rule_dpif_get_flow_cookie(const struct rule_dpif
*rule
)
3171 OVS_REQUIRES(rule
->up
.mutex
)
3173 return rule
->up
.flow_cookie
;
3177 rule_dpif_reduce_timeouts(struct rule_dpif
*rule
, uint16_t idle_timeout
,
3178 uint16_t hard_timeout
)
3180 ofproto_rule_reduce_timeouts(&rule
->up
, idle_timeout
, hard_timeout
);
3183 /* Returns 'rule''s actions. The caller owns a reference on the returned
3184 * actions and must eventually release it (with rule_actions_unref()) to avoid
3186 const struct rule_actions
*
3187 rule_dpif_get_actions(const struct rule_dpif
*rule
)
3189 return rule_get_actions(&rule
->up
);
3192 /* Lookup 'flow' in table 0 of 'ofproto''s classifier.
3193 * If 'wc' is non-null, sets the fields that were relevant as part of
3194 * the lookup. Returns the table_id where a match or miss occurred.
3196 * The return value will be zero unless there was a miss and
3197 * OFPTC11_TABLE_MISS_CONTINUE is in effect for the sequence of tables
3198 * where misses occur.
3200 * The rule is returned in '*rule', which is valid at least until the next
3201 * RCU quiescent period. If the '*rule' needs to stay around longer,
3202 * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3203 * on it before this returns. */
3205 rule_dpif_lookup(struct ofproto_dpif
*ofproto
, struct flow
*flow
,
3206 struct flow_wildcards
*wc
, struct rule_dpif
**rule
,
3209 enum rule_dpif_lookup_verdict verdict
;
3210 enum ofputil_port_config config
= 0;
3213 if (ofproto_dpif_get_enable_recirc(ofproto
)) {
3214 /* Always exactly match recirc_id since datapath supports
3217 wc
->masks
.recirc_id
= UINT32_MAX
;
3220 /* Start looking up from internal table for post recirculation flows
3221 * or packets. We can also simply send all, including normal flows
3222 * or packets to the internal table. They will not match any post
3223 * recirculation rules except the 'catch all' rule that resubmit
3226 * As an optimization, we send normal flows and packets to table 0
3227 * directly, saving one table lookup. */
3228 table_id
= flow
->recirc_id
? TBL_INTERNAL
: 0;
3233 verdict
= rule_dpif_lookup_from_table(ofproto
, flow
, wc
, true,
3234 &table_id
, rule
, take_ref
);
3237 case RULE_DPIF_LOOKUP_VERDICT_MATCH
:
3239 case RULE_DPIF_LOOKUP_VERDICT_CONTROLLER
: {
3240 struct ofport_dpif
*port
;
3242 port
= get_ofp_port(ofproto
, flow
->in_port
.ofp_port
);
3244 VLOG_WARN_RL(&rl
, "packet-in on unknown OpenFlow port %"PRIu16
,
3245 flow
->in_port
.ofp_port
);
3247 config
= port
? port
->up
.pp
.config
: 0;
3250 case RULE_DPIF_LOOKUP_VERDICT_DROP
:
3251 config
= OFPUTIL_PC_NO_PACKET_IN
;
3253 case RULE_DPIF_LOOKUP_VERDICT_DEFAULT
:
3254 if (!connmgr_wants_packet_in_on_miss(ofproto
->up
.connmgr
)) {
3255 config
= OFPUTIL_PC_NO_PACKET_IN
;
3262 choose_miss_rule(config
, ofproto
->miss_rule
,
3263 ofproto
->no_packet_in_rule
, rule
, take_ref
);
3267 /* The returned rule is valid at least until the next RCU quiescent period.
3268 * If the '*rule' needs to stay around longer, a non-zero 'take_ref' must be
3269 * passed in to cause a reference to be taken on it before this returns. */
3270 static struct rule_dpif
*
3271 rule_dpif_lookup_in_table(struct ofproto_dpif
*ofproto
, uint8_t table_id
,
3272 const struct flow
*flow
, struct flow_wildcards
*wc
,
3275 struct classifier
*cls
= &ofproto
->up
.tables
[table_id
].cls
;
3276 const struct cls_rule
*cls_rule
;
3277 struct rule_dpif
*rule
;
3279 fat_rwlock_rdlock(&cls
->rwlock
);
3280 if (ofproto
->up
.frag_handling
!= OFPC_FRAG_NX_MATCH
) {
3282 memset(&wc
->masks
.dl_type
, 0xff, sizeof wc
->masks
.dl_type
);
3283 if (is_ip_any(flow
)) {
3284 wc
->masks
.nw_frag
|= FLOW_NW_FRAG_MASK
;
3288 if (flow
->nw_frag
& FLOW_NW_FRAG_ANY
) {
3289 if (ofproto
->up
.frag_handling
== OFPC_FRAG_NORMAL
) {
3290 /* We must pretend that transport ports are unavailable. */
3291 struct flow ofpc_normal_flow
= *flow
;
3292 ofpc_normal_flow
.tp_src
= htons(0);
3293 ofpc_normal_flow
.tp_dst
= htons(0);
3294 cls_rule
= classifier_lookup(cls
, &ofpc_normal_flow
, wc
);
3296 /* Must be OFPC_FRAG_DROP (we don't have OFPC_FRAG_REASM). */
3297 cls_rule
= &ofproto
->drop_frags_rule
->up
.cr
;
3300 cls_rule
= classifier_lookup(cls
, flow
, wc
);
3303 cls_rule
= classifier_lookup(cls
, flow
, wc
);
3306 rule
= rule_dpif_cast(rule_from_cls_rule(cls_rule
));
3308 rule_dpif_ref(rule
);
3310 fat_rwlock_unlock(&cls
->rwlock
);
3315 /* Look up 'flow' in 'ofproto''s classifier starting from table '*table_id'.
3316 * Stores the rule that was found in '*rule', or NULL if none was found.
3317 * Updates 'wc', if nonnull, to reflect the fields that were used during the
3320 * If 'honor_table_miss' is true, the first lookup occurs in '*table_id', but
3321 * if none is found then the table miss configuration for that table is
3322 * honored, which can result in additional lookups in other OpenFlow tables.
3323 * In this case the function updates '*table_id' to reflect the final OpenFlow
3324 * table that was searched.
3326 * If 'honor_table_miss' is false, then only one table lookup occurs, in
3331 * - RULE_DPIF_LOOKUP_VERDICT_MATCH if a rule (in '*rule') was found.
3333 * - RULE_OFPTC_TABLE_MISS_CONTROLLER if no rule was found and either:
3334 * + 'honor_table_miss' is false
3335 * + a table miss configuration specified that the packet should be
3336 * sent to the controller in this case.
3338 * - RULE_DPIF_LOOKUP_VERDICT_DROP if no rule was found, 'honor_table_miss'
3339 * is true and a table miss configuration specified that the packet
3340 * should be dropped in this case.
3342 * - RULE_DPIF_LOOKUP_VERDICT_DEFAULT if no rule was found,
3343 * 'honor_table_miss' is true and a table miss configuration has
3344 * not been specified in this case.
3346 * The rule is returned in '*rule', which is valid at least until the next
3347 * RCU quiescent period. If the '*rule' needs to stay around longer,
3348 * a non-zero 'take_ref' must be passed in to cause a reference to be taken
3349 * on it before this returns. */
3350 enum rule_dpif_lookup_verdict
3351 rule_dpif_lookup_from_table(struct ofproto_dpif
*ofproto
,
3352 const struct flow
*flow
,
3353 struct flow_wildcards
*wc
,
3354 bool honor_table_miss
,
3355 uint8_t *table_id
, struct rule_dpif
**rule
,
3360 for (next_id
= *table_id
;
3361 next_id
< ofproto
->up
.n_tables
;
3362 next_id
++, next_id
+= (next_id
== TBL_INTERNAL
))
3364 *table_id
= next_id
;
3365 *rule
= rule_dpif_lookup_in_table(ofproto
, *table_id
, flow
, wc
,
3368 return RULE_DPIF_LOOKUP_VERDICT_MATCH
;
3369 } else if (!honor_table_miss
) {
3370 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER
;
3372 switch (ofproto_table_get_config(&ofproto
->up
, *table_id
)) {
3373 case OFPROTO_TABLE_MISS_CONTINUE
:
3376 case OFPROTO_TABLE_MISS_CONTROLLER
:
3377 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER
;
3379 case OFPROTO_TABLE_MISS_DROP
:
3380 return RULE_DPIF_LOOKUP_VERDICT_DROP
;
3382 case OFPROTO_TABLE_MISS_DEFAULT
:
3383 return RULE_DPIF_LOOKUP_VERDICT_DEFAULT
;
3388 return RULE_DPIF_LOOKUP_VERDICT_CONTROLLER
;
3391 /* Given a port configuration (specified as zero if there's no port), chooses
3392 * which of 'miss_rule' and 'no_packet_in_rule' should be used in case of a
3395 * The rule is returned in '*rule', which is valid at least until the next
3396 * RCU quiescent period. If the '*rule' needs to stay around longer,
3397 * a reference must be taken on it (rule_dpif_ref()).
3400 choose_miss_rule(enum ofputil_port_config config
, struct rule_dpif
*miss_rule
,
3401 struct rule_dpif
*no_packet_in_rule
, struct rule_dpif
**rule
,
3404 *rule
= config
& OFPUTIL_PC_NO_PACKET_IN
? no_packet_in_rule
: miss_rule
;
3406 rule_dpif_ref(*rule
);
3411 complete_operation(struct rule_dpif
*rule
)
3412 OVS_REQUIRES(ofproto_mutex
)
3414 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
3416 ofproto
->backer
->need_revalidate
= REV_FLOW_TABLE
;
3417 ofoperation_complete(rule
->up
.pending
, 0);
3420 static struct rule_dpif
*rule_dpif_cast(const struct rule
*rule
)
3422 return rule
? CONTAINER_OF(rule
, struct rule_dpif
, up
) : NULL
;
3425 static struct rule
*
3428 struct rule_dpif
*rule
= xmalloc(sizeof *rule
);
3433 rule_dealloc(struct rule
*rule_
)
3435 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
3440 rule_construct(struct rule
*rule_
)
3441 OVS_NO_THREAD_SAFETY_ANALYSIS
3443 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
3444 ovs_mutex_init_adaptive(&rule
->stats_mutex
);
3445 rule
->stats
.n_packets
= 0;
3446 rule
->stats
.n_bytes
= 0;
3447 rule
->stats
.used
= rule
->up
.modified
;
3452 rule_insert(struct rule
*rule_
)
3453 OVS_REQUIRES(ofproto_mutex
)
3455 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
3456 complete_operation(rule
);
3460 rule_delete(struct rule
*rule_
)
3461 OVS_REQUIRES(ofproto_mutex
)
3463 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
3464 complete_operation(rule
);
3468 rule_destruct(struct rule
*rule_
)
3470 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
3471 ovs_mutex_destroy(&rule
->stats_mutex
);
3475 rule_get_stats(struct rule
*rule_
, uint64_t *packets
, uint64_t *bytes
,
3476 long long int *used
)
3478 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
3480 ovs_mutex_lock(&rule
->stats_mutex
);
3481 *packets
= rule
->stats
.n_packets
;
3482 *bytes
= rule
->stats
.n_bytes
;
3483 *used
= rule
->stats
.used
;
3484 ovs_mutex_unlock(&rule
->stats_mutex
);
3488 rule_dpif_execute(struct rule_dpif
*rule
, const struct flow
*flow
,
3489 struct ofpbuf
*packet
)
3491 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(rule
->up
.ofproto
);
3493 ofproto_dpif_execute_actions(ofproto
, flow
, rule
, NULL
, 0, packet
);
3497 rule_execute(struct rule
*rule
, const struct flow
*flow
,
3498 struct ofpbuf
*packet
)
3500 rule_dpif_execute(rule_dpif_cast(rule
), flow
, packet
);
3501 ofpbuf_delete(packet
);
3506 rule_modify_actions(struct rule
*rule_
, bool reset_counters
)
3507 OVS_REQUIRES(ofproto_mutex
)
3509 struct rule_dpif
*rule
= rule_dpif_cast(rule_
);
3511 if (reset_counters
) {
3512 ovs_mutex_lock(&rule
->stats_mutex
);
3513 rule
->stats
.n_packets
= 0;
3514 rule
->stats
.n_bytes
= 0;
3515 ovs_mutex_unlock(&rule
->stats_mutex
);
3518 complete_operation(rule
);
3521 static struct group_dpif
*group_dpif_cast(const struct ofgroup
*group
)
3523 return group
? CONTAINER_OF(group
, struct group_dpif
, up
) : NULL
;
3526 static struct ofgroup
*
3529 struct group_dpif
*group
= xzalloc(sizeof *group
);
3534 group_dealloc(struct ofgroup
*group_
)
3536 struct group_dpif
*group
= group_dpif_cast(group_
);
3541 group_construct_stats(struct group_dpif
*group
)
3542 OVS_REQUIRES(group
->stats_mutex
)
3544 group
->packet_count
= 0;
3545 group
->byte_count
= 0;
3546 if (!group
->bucket_stats
) {
3547 group
->bucket_stats
= xcalloc(group
->up
.n_buckets
,
3548 sizeof *group
->bucket_stats
);
3550 memset(group
->bucket_stats
, 0, group
->up
.n_buckets
*
3551 sizeof *group
->bucket_stats
);
3556 group_construct(struct ofgroup
*group_
)
3558 struct group_dpif
*group
= group_dpif_cast(group_
);
3559 const struct ofputil_bucket
*bucket
;
3561 /* Prevent group chaining because our locking structure makes it hard to
3562 * implement deadlock-free. (See xlate_group_resource_check().) */
3563 LIST_FOR_EACH (bucket
, list_node
, &group
->up
.buckets
) {
3564 const struct ofpact
*a
;
3566 OFPACT_FOR_EACH (a
, bucket
->ofpacts
, bucket
->ofpacts_len
) {
3567 if (a
->type
== OFPACT_GROUP
) {
3568 return OFPERR_OFPGMFC_CHAINING_UNSUPPORTED
;
3573 ovs_mutex_init_adaptive(&group
->stats_mutex
);
3574 ovs_mutex_lock(&group
->stats_mutex
);
3575 group_construct_stats(group
);
3576 ovs_mutex_unlock(&group
->stats_mutex
);
3581 group_destruct__(struct group_dpif
*group
)
3582 OVS_REQUIRES(group
->stats_mutex
)
3584 free(group
->bucket_stats
);
3585 group
->bucket_stats
= NULL
;
3589 group_destruct(struct ofgroup
*group_
)
3591 struct group_dpif
*group
= group_dpif_cast(group_
);
3592 ovs_mutex_lock(&group
->stats_mutex
);
3593 group_destruct__(group
);
3594 ovs_mutex_unlock(&group
->stats_mutex
);
3595 ovs_mutex_destroy(&group
->stats_mutex
);
3599 group_modify(struct ofgroup
*group_
, struct ofgroup
*victim_
)
3601 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(group_
->ofproto
);
3602 struct group_dpif
*group
= group_dpif_cast(group_
);
3603 struct group_dpif
*victim
= group_dpif_cast(victim_
);
3605 ovs_mutex_lock(&group
->stats_mutex
);
3606 if (victim
->up
.n_buckets
< group
->up
.n_buckets
) {
3607 group_destruct__(group
);
3609 group_construct_stats(group
);
3610 ovs_mutex_unlock(&group
->stats_mutex
);
3612 ofproto
->backer
->need_revalidate
= REV_FLOW_TABLE
;
3618 group_get_stats(const struct ofgroup
*group_
, struct ofputil_group_stats
*ogs
)
3620 struct group_dpif
*group
= group_dpif_cast(group_
);
3622 ovs_mutex_lock(&group
->stats_mutex
);
3623 ogs
->packet_count
= group
->packet_count
;
3624 ogs
->byte_count
= group
->byte_count
;
3625 memcpy(ogs
->bucket_stats
, group
->bucket_stats
,
3626 group
->up
.n_buckets
* sizeof *group
->bucket_stats
);
3627 ovs_mutex_unlock(&group
->stats_mutex
);
3633 group_dpif_lookup(struct ofproto_dpif
*ofproto
, uint32_t group_id
,
3634 struct group_dpif
**group
)
3635 OVS_TRY_RDLOCK(true, (*group
)->up
.rwlock
)
3637 struct ofgroup
*ofgroup
;
3641 found
= ofproto_group_lookup(&ofproto
->up
, group_id
, &ofgroup
);
3642 *group
= found
? group_dpif_cast(ofgroup
) : NULL
;
3648 group_dpif_release(struct group_dpif
*group
)
3649 OVS_RELEASES(group
->up
.rwlock
)
3651 ofproto_group_release(&group
->up
);
3655 group_dpif_get_buckets(const struct group_dpif
*group
,
3656 const struct list
**buckets
)
3658 *buckets
= &group
->up
.buckets
;
3661 enum ofp11_group_type
3662 group_dpif_get_type(const struct group_dpif
*group
)
3664 return group
->up
.type
;
3667 /* Sends 'packet' out 'ofport'.
3668 * May modify 'packet'.
3669 * Returns 0 if successful, otherwise a positive errno value. */
3671 ofproto_dpif_send_packet(const struct ofport_dpif
*ofport
, struct ofpbuf
*packet
)
3673 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport
->up
.ofproto
);
3676 error
= xlate_send_packet(ofport
, packet
);
3678 ovs_mutex_lock(&ofproto
->stats_mutex
);
3679 ofproto
->stats
.tx_packets
++;
3680 ofproto
->stats
.tx_bytes
+= ofpbuf_size(packet
);
3681 ovs_mutex_unlock(&ofproto
->stats_mutex
);
3686 set_frag_handling(struct ofproto
*ofproto_
,
3687 enum ofp_config_flags frag_handling
)
3689 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3690 if (frag_handling
!= OFPC_FRAG_REASM
) {
3691 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3699 packet_out(struct ofproto
*ofproto_
, struct ofpbuf
*packet
,
3700 const struct flow
*flow
,
3701 const struct ofpact
*ofpacts
, size_t ofpacts_len
)
3703 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3705 ofproto_dpif_execute_actions(ofproto
, flow
, NULL
, ofpacts
,
3706 ofpacts_len
, packet
);
3713 set_netflow(struct ofproto
*ofproto_
,
3714 const struct netflow_options
*netflow_options
)
3716 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3718 if (netflow_options
) {
3719 if (!ofproto
->netflow
) {
3720 ofproto
->netflow
= netflow_create();
3721 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3723 return netflow_set_options(ofproto
->netflow
, netflow_options
);
3724 } else if (ofproto
->netflow
) {
3725 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
3726 netflow_unref(ofproto
->netflow
);
3727 ofproto
->netflow
= NULL
;
3734 get_netflow_ids(const struct ofproto
*ofproto_
,
3735 uint8_t *engine_type
, uint8_t *engine_id
)
3737 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofproto_
);
3739 dpif_get_netflow_ids(ofproto
->backer
->dpif
, engine_type
, engine_id
);
3742 static struct ofproto_dpif
*
3743 ofproto_dpif_lookup(const char *name
)
3745 struct ofproto_dpif
*ofproto
;
3747 HMAP_FOR_EACH_WITH_HASH (ofproto
, all_ofproto_dpifs_node
,
3748 hash_string(name
, 0), &all_ofproto_dpifs
) {
3749 if (!strcmp(ofproto
->up
.name
, name
)) {
3757 ofproto_unixctl_fdb_flush(struct unixctl_conn
*conn
, int argc
,
3758 const char *argv
[], void *aux OVS_UNUSED
)
3760 struct ofproto_dpif
*ofproto
;
3763 ofproto
= ofproto_dpif_lookup(argv
[1]);
3765 unixctl_command_reply_error(conn
, "no such bridge");
3768 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
3769 mac_learning_flush(ofproto
->ml
);
3770 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
3772 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
3773 ovs_rwlock_wrlock(&ofproto
->ml
->rwlock
);
3774 mac_learning_flush(ofproto
->ml
);
3775 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
3779 unixctl_command_reply(conn
, "table successfully flushed");
3782 static struct ofport_dpif
*
3783 ofbundle_get_a_port(const struct ofbundle
*bundle
)
3785 return CONTAINER_OF(list_front(&bundle
->ports
), struct ofport_dpif
,
3790 ofproto_unixctl_fdb_show(struct unixctl_conn
*conn
, int argc OVS_UNUSED
,
3791 const char *argv
[], void *aux OVS_UNUSED
)
3793 struct ds ds
= DS_EMPTY_INITIALIZER
;
3794 const struct ofproto_dpif
*ofproto
;
3795 const struct mac_entry
*e
;
3797 ofproto
= ofproto_dpif_lookup(argv
[1]);
3799 unixctl_command_reply_error(conn
, "no such bridge");
3803 ds_put_cstr(&ds
, " port VLAN MAC Age\n");
3804 ovs_rwlock_rdlock(&ofproto
->ml
->rwlock
);
3805 LIST_FOR_EACH (e
, lru_node
, &ofproto
->ml
->lrus
) {
3806 struct ofbundle
*bundle
= e
->port
.p
;
3807 char name
[OFP_MAX_PORT_NAME_LEN
];
3809 ofputil_port_to_string(ofbundle_get_a_port(bundle
)->up
.ofp_port
,
3811 ds_put_format(&ds
, "%5s %4d "ETH_ADDR_FMT
" %3d\n",
3812 name
, e
->vlan
, ETH_ADDR_ARGS(e
->mac
),
3813 mac_entry_age(ofproto
->ml
, e
));
3815 ovs_rwlock_unlock(&ofproto
->ml
->rwlock
);
3816 unixctl_command_reply(conn
, ds_cstr(&ds
));
3821 struct xlate_out xout
;
3822 struct xlate_in xin
;
3823 const struct flow
*key
;
3825 struct flow_wildcards wc
;
3830 trace_format_rule(struct ds
*result
, int level
, const struct rule_dpif
*rule
)
3832 const struct rule_actions
*actions
;
3835 ds_put_char_multiple(result
, '\t', level
);
3837 ds_put_cstr(result
, "No match\n");
3841 ovs_mutex_lock(&rule
->up
.mutex
);
3842 cookie
= rule
->up
.flow_cookie
;
3843 ovs_mutex_unlock(&rule
->up
.mutex
);
3845 ds_put_format(result
, "Rule: table=%"PRIu8
" cookie=%#"PRIx64
" ",
3846 rule
? rule
->up
.table_id
: 0, ntohll(cookie
));
3847 cls_rule_format(&rule
->up
.cr
, result
);
3848 ds_put_char(result
, '\n');
3850 actions
= rule_dpif_get_actions(rule
);
3852 ds_put_char_multiple(result
, '\t', level
);
3853 ds_put_cstr(result
, "OpenFlow actions=");
3854 ofpacts_format(actions
->ofpacts
, actions
->ofpacts_len
, result
);
3855 ds_put_char(result
, '\n');
3859 trace_format_flow(struct ds
*result
, int level
, const char *title
,
3860 struct trace_ctx
*trace
)
3862 ds_put_char_multiple(result
, '\t', level
);
3863 ds_put_format(result
, "%s: ", title
);
3864 /* Do not report unchanged flows for resubmits. */
3865 if ((level
> 0 && flow_equal(&trace
->xin
.flow
, &trace
->flow
))
3866 || (level
== 0 && flow_equal(&trace
->xin
.flow
, trace
->key
))) {
3867 ds_put_cstr(result
, "unchanged");
3869 flow_format(result
, &trace
->xin
.flow
);
3870 trace
->flow
= trace
->xin
.flow
;
3872 ds_put_char(result
, '\n');
3876 trace_format_regs(struct ds
*result
, int level
, const char *title
,
3877 struct trace_ctx
*trace
)
3881 ds_put_char_multiple(result
, '\t', level
);
3882 ds_put_format(result
, "%s:", title
);
3883 for (i
= 0; i
< FLOW_N_REGS
; i
++) {
3884 ds_put_format(result
, " reg%"PRIuSIZE
"=0x%"PRIx32
, i
, trace
->flow
.regs
[i
]);
3886 ds_put_char(result
, '\n');
3890 trace_format_odp(struct ds
*result
, int level
, const char *title
,
3891 struct trace_ctx
*trace
)
3893 struct ofpbuf
*odp_actions
= &trace
->xout
.odp_actions
;
3895 ds_put_char_multiple(result
, '\t', level
);
3896 ds_put_format(result
, "%s: ", title
);
3897 format_odp_actions(result
, ofpbuf_data(odp_actions
),
3898 ofpbuf_size(odp_actions
));
3899 ds_put_char(result
, '\n');
3903 trace_format_megaflow(struct ds
*result
, int level
, const char *title
,
3904 struct trace_ctx
*trace
)
3908 ds_put_char_multiple(result
, '\t', level
);
3909 ds_put_format(result
, "%s: ", title
);
3910 flow_wildcards_or(&trace
->wc
, &trace
->xout
.wc
, &trace
->wc
);
3911 match_init(&match
, trace
->key
, &trace
->wc
);
3912 match_format(&match
, result
, OFP_DEFAULT_PRIORITY
);
3913 ds_put_char(result
, '\n');
3917 trace_resubmit(struct xlate_in
*xin
, struct rule_dpif
*rule
, int recurse
)
3919 struct trace_ctx
*trace
= CONTAINER_OF(xin
, struct trace_ctx
, xin
);
3920 struct ds
*result
= trace
->result
;
3922 ds_put_char(result
, '\n');
3923 trace_format_flow(result
, recurse
+ 1, "Resubmitted flow", trace
);
3924 trace_format_regs(result
, recurse
+ 1, "Resubmitted regs", trace
);
3925 trace_format_odp(result
, recurse
+ 1, "Resubmitted odp", trace
);
3926 trace_format_megaflow(result
, recurse
+ 1, "Resubmitted megaflow", trace
);
3927 trace_format_rule(result
, recurse
+ 1, rule
);
3931 trace_report(struct xlate_in
*xin
, const char *s
, int recurse
)
3933 struct trace_ctx
*trace
= CONTAINER_OF(xin
, struct trace_ctx
, xin
);
3934 struct ds
*result
= trace
->result
;
3936 ds_put_char_multiple(result
, '\t', recurse
);
3937 ds_put_cstr(result
, s
);
3938 ds_put_char(result
, '\n');
3941 /* Parses the 'argc' elements of 'argv', ignoring argv[0]. The following
3942 * forms are supported:
3944 * - [dpname] odp_flow [-generate | packet]
3945 * - bridge br_flow [-generate | packet]
3947 * On success, initializes '*ofprotop' and 'flow' and returns NULL. On failure
3948 * returns a nonnull malloced error message. */
3949 static char * WARN_UNUSED_RESULT
3950 parse_flow_and_packet(int argc
, const char *argv
[],
3951 struct ofproto_dpif
**ofprotop
, struct flow
*flow
,
3952 struct ofpbuf
**packetp
)
3954 const struct dpif_backer
*backer
= NULL
;
3955 const char *error
= NULL
;
3957 struct simap port_names
= SIMAP_INITIALIZER(&port_names
);
3958 struct ofpbuf
*packet
;
3959 struct ofpbuf odp_key
;
3960 struct ofpbuf odp_mask
;
3962 ofpbuf_init(&odp_key
, 0);
3963 ofpbuf_init(&odp_mask
, 0);
3965 /* Handle "-generate" or a hex string as the last argument. */
3966 if (!strcmp(argv
[argc
- 1], "-generate")) {
3967 packet
= ofpbuf_new(0);
3970 error
= eth_from_hex(argv
[argc
- 1], &packet
);
3973 } else if (argc
== 4) {
3974 /* The 3-argument form must end in "-generate' or a hex string. */
3980 /* odp_flow can have its in_port specified as a name instead of port no.
3981 * We do not yet know whether a given flow is a odp_flow or a br_flow.
3982 * But, to know whether a flow is odp_flow through odp_flow_from_string(),
3983 * we need to create a simap of name to port no. */
3985 const char *dp_type
;
3986 if (!strncmp(argv
[1], "ovs-", 4)) {
3987 dp_type
= argv
[1] + 4;
3991 backer
= shash_find_data(&all_dpif_backers
, dp_type
);
3992 } else if (argc
== 2) {
3993 struct shash_node
*node
;
3994 if (shash_count(&all_dpif_backers
) == 1) {
3995 node
= shash_first(&all_dpif_backers
);
3996 backer
= node
->data
;
3999 error
= "Syntax error";
4002 if (backer
&& backer
->dpif
) {
4003 struct dpif_port dpif_port
;
4004 struct dpif_port_dump port_dump
;
4005 DPIF_PORT_FOR_EACH (&dpif_port
, &port_dump
, backer
->dpif
) {
4006 simap_put(&port_names
, dpif_port
.name
,
4007 odp_to_u32(dpif_port
.port_no
));
4011 /* Parse the flow and determine whether a datapath or
4012 * bridge is specified. If function odp_flow_key_from_string()
4013 * returns 0, the flow is a odp_flow. If function
4014 * parse_ofp_exact_flow() returns NULL, the flow is a br_flow. */
4015 if (!odp_flow_from_string(argv
[argc
- 1], &port_names
,
4016 &odp_key
, &odp_mask
)) {
4018 error
= "Cannot find the datapath";
4022 if (xlate_receive(backer
, NULL
, ofpbuf_data(&odp_key
),
4023 ofpbuf_size(&odp_key
), flow
,
4024 ofprotop
, NULL
, NULL
, NULL
, NULL
)) {
4025 error
= "Invalid datapath flow";
4029 char *err
= parse_ofp_exact_flow(flow
, NULL
, argv
[argc
- 1], NULL
);
4032 m_err
= xasprintf("Bad flow syntax: %s", err
);
4037 error
= "Must specify bridge name";
4041 *ofprotop
= ofproto_dpif_lookup(argv
[1]);
4043 error
= "Unknown bridge name";
4049 /* Generate a packet, if requested. */
4051 if (!ofpbuf_size(packet
)) {
4052 flow_compose(packet
, flow
);
4054 struct pkt_metadata md
= pkt_metadata_from_flow(flow
);
4056 /* Use the metadata from the flow and the packet argument
4057 * to reconstruct the flow. */
4058 flow_extract(packet
, &md
, flow
);
4063 if (error
&& !m_err
) {
4064 m_err
= xstrdup(error
);
4067 ofpbuf_delete(packet
);
4071 ofpbuf_uninit(&odp_key
);
4072 ofpbuf_uninit(&odp_mask
);
4073 simap_destroy(&port_names
);
4078 ofproto_unixctl_trace(struct unixctl_conn
*conn
, int argc
, const char *argv
[],
4079 void *aux OVS_UNUSED
)
4081 struct ofproto_dpif
*ofproto
;
4082 struct ofpbuf
*packet
;
4086 error
= parse_flow_and_packet(argc
, argv
, &ofproto
, &flow
, &packet
);
4091 ofproto_trace(ofproto
, &flow
, packet
, NULL
, 0, &result
);
4092 unixctl_command_reply(conn
, ds_cstr(&result
));
4093 ds_destroy(&result
);
4094 ofpbuf_delete(packet
);
4096 unixctl_command_reply_error(conn
, error
);
4102 ofproto_unixctl_trace_actions(struct unixctl_conn
*conn
, int argc
,
4103 const char *argv
[], void *aux OVS_UNUSED
)
4105 enum ofputil_protocol usable_protocols
;
4106 struct ofproto_dpif
*ofproto
;
4107 bool enforce_consistency
;
4108 struct ofpbuf ofpacts
;
4109 struct ofpbuf
*packet
;
4114 /* Three kinds of error return values! */
4120 ofpbuf_init(&ofpacts
, 0);
4122 /* Parse actions. */
4123 error
= parse_ofpacts(argv
[--argc
], &ofpacts
, &usable_protocols
);
4125 unixctl_command_reply_error(conn
, error
);
4130 /* OpenFlow 1.1 and later suggest that the switch enforces certain forms of
4131 * consistency between the flow and the actions. With -consistent, we
4132 * enforce consistency even for a flow supported in OpenFlow 1.0. */
4133 if (!strcmp(argv
[1], "-consistent")) {
4134 enforce_consistency
= true;
4138 enforce_consistency
= false;
4141 error
= parse_flow_and_packet(argc
, argv
, &ofproto
, &flow
, &packet
);
4143 unixctl_command_reply_error(conn
, error
);
4148 /* Do the same checks as handle_packet_out() in ofproto.c.
4150 * We pass a 'table_id' of 0 to ofproto_check_ofpacts(), which isn't
4151 * strictly correct because these actions aren't in any table, but it's OK
4152 * because it 'table_id' is used only to check goto_table instructions, but
4153 * packet-outs take a list of actions and therefore it can't include
4156 * We skip the "meter" check here because meter is an instruction, not an
4157 * action, and thus cannot appear in ofpacts. */
4158 in_port
= ofp_to_u16(flow
.in_port
.ofp_port
);
4159 if (in_port
>= ofproto
->up
.max_ports
&& in_port
< ofp_to_u16(OFPP_MAX
)) {
4160 unixctl_command_reply_error(conn
, "invalid in_port");
4163 if (enforce_consistency
) {
4164 retval
= ofpacts_check_consistency(ofpbuf_data(&ofpacts
), ofpbuf_size(&ofpacts
),
4165 &flow
, u16_to_ofp(ofproto
->up
.max_ports
),
4166 0, 0, usable_protocols
);
4168 retval
= ofpacts_check(ofpbuf_data(&ofpacts
), ofpbuf_size(&ofpacts
), &flow
,
4169 u16_to_ofp(ofproto
->up
.max_ports
), 0, 0,
4175 ds_put_format(&result
, "Bad actions: %s", ofperr_to_string(retval
));
4176 unixctl_command_reply_error(conn
, ds_cstr(&result
));
4180 ofproto_trace(ofproto
, &flow
, packet
,
4181 ofpbuf_data(&ofpacts
), ofpbuf_size(&ofpacts
), &result
);
4182 unixctl_command_reply(conn
, ds_cstr(&result
));
4185 ds_destroy(&result
);
4186 ofpbuf_delete(packet
);
4187 ofpbuf_uninit(&ofpacts
);
4190 /* Implements a "trace" through 'ofproto''s flow table, appending a textual
4191 * description of the results to 'ds'.
4193 * The trace follows a packet with the specified 'flow' through the flow
4194 * table. 'packet' may be nonnull to trace an actual packet, with consequent
4195 * side effects (if it is nonnull then its flow must be 'flow').
4197 * If 'ofpacts' is nonnull then its 'ofpacts_len' bytes specify the actions to
4198 * trace, otherwise the actions are determined by a flow table lookup. */
4200 ofproto_trace(struct ofproto_dpif
*ofproto
, struct flow
*flow
,
4201 const struct ofpbuf
*packet
,
4202 const struct ofpact ofpacts
[], size_t ofpacts_len
,
4205 struct rule_dpif
*rule
;
4206 struct trace_ctx trace
;
4208 ds_put_format(ds
, "Bridge: %s\n", ofproto
->up
.name
);
4209 ds_put_cstr(ds
, "Flow: ");
4210 flow_format(ds
, flow
);
4211 ds_put_char(ds
, '\n');
4213 flow_wildcards_init_catchall(&trace
.wc
);
4217 rule_dpif_lookup(ofproto
, flow
, &trace
.wc
, &rule
, false);
4219 trace_format_rule(ds
, 0, rule
);
4220 if (rule
== ofproto
->miss_rule
) {
4221 ds_put_cstr(ds
, "\nNo match, flow generates \"packet in\"s.\n");
4222 } else if (rule
== ofproto
->no_packet_in_rule
) {
4223 ds_put_cstr(ds
, "\nNo match, packets dropped because "
4224 "OFPPC_NO_PACKET_IN is set on in_port.\n");
4225 } else if (rule
== ofproto
->drop_frags_rule
) {
4226 ds_put_cstr(ds
, "\nPackets dropped because they are IP fragments "
4227 "and the fragment handling mode is \"drop\".\n");
4231 if (rule
|| ofpacts
) {
4233 trace
.key
= flow
; /* Original flow key, used for megaflow. */
4234 trace
.flow
= *flow
; /* May be modified by actions. */
4235 xlate_in_init(&trace
.xin
, ofproto
, flow
, rule
, ntohs(flow
->tcp_flags
),
4238 trace
.xin
.ofpacts
= ofpacts
;
4239 trace
.xin
.ofpacts_len
= ofpacts_len
;
4241 trace
.xin
.resubmit_hook
= trace_resubmit
;
4242 trace
.xin
.report_hook
= trace_report
;
4244 xlate_actions(&trace
.xin
, &trace
.xout
);
4246 ds_put_char(ds
, '\n');
4247 trace_format_flow(ds
, 0, "Final flow", &trace
);
4248 trace_format_megaflow(ds
, 0, "Megaflow", &trace
);
4250 ds_put_cstr(ds
, "Datapath actions: ");
4251 format_odp_actions(ds
, ofpbuf_data(&trace
.xout
.odp_actions
),
4252 ofpbuf_size(&trace
.xout
.odp_actions
));
4254 if (trace
.xout
.slow
) {
4255 enum slow_path_reason slow
;
4257 ds_put_cstr(ds
, "\nThis flow is handled by the userspace "
4258 "slow path because it:");
4260 slow
= trace
.xout
.slow
;
4262 enum slow_path_reason bit
= rightmost_1bit(slow
);
4264 ds_put_format(ds
, "\n\t- %s.",
4265 slow_path_reason_to_explanation(bit
));
4271 xlate_out_uninit(&trace
.xout
);
4275 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
4276 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
4277 * to destroy 'ofproto_shash' and free the returned value. */
4278 static const struct shash_node
**
4279 get_ofprotos(struct shash
*ofproto_shash
)
4281 const struct ofproto_dpif
*ofproto
;
4283 HMAP_FOR_EACH (ofproto
, all_ofproto_dpifs_node
, &all_ofproto_dpifs
) {
4284 char *name
= xasprintf("%s@%s", ofproto
->up
.type
, ofproto
->up
.name
);
4285 shash_add_nocopy(ofproto_shash
, name
, ofproto
);
4288 return shash_sort(ofproto_shash
);
4292 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn
*conn
, int argc OVS_UNUSED
,
4293 const char *argv
[] OVS_UNUSED
,
4294 void *aux OVS_UNUSED
)
4296 struct ds ds
= DS_EMPTY_INITIALIZER
;
4297 struct shash ofproto_shash
;
4298 const struct shash_node
**sorted_ofprotos
;
4301 shash_init(&ofproto_shash
);
4302 sorted_ofprotos
= get_ofprotos(&ofproto_shash
);
4303 for (i
= 0; i
< shash_count(&ofproto_shash
); i
++) {
4304 const struct shash_node
*node
= sorted_ofprotos
[i
];
4305 ds_put_format(&ds
, "%s\n", node
->name
);
4308 shash_destroy(&ofproto_shash
);
4309 free(sorted_ofprotos
);
4311 unixctl_command_reply(conn
, ds_cstr(&ds
));
4316 dpif_show_backer(const struct dpif_backer
*backer
, struct ds
*ds
)
4318 const struct shash_node
**ofprotos
;
4319 struct dpif_dp_stats dp_stats
;
4320 struct shash ofproto_shash
;
4323 dpif_get_dp_stats(backer
->dpif
, &dp_stats
);
4325 ds_put_format(ds
, "%s: hit:%"PRIu64
" missed:%"PRIu64
"\n",
4326 dpif_name(backer
->dpif
), dp_stats
.n_hit
, dp_stats
.n_missed
);
4328 shash_init(&ofproto_shash
);
4329 ofprotos
= get_ofprotos(&ofproto_shash
);
4330 for (i
= 0; i
< shash_count(&ofproto_shash
); i
++) {
4331 struct ofproto_dpif
*ofproto
= ofprotos
[i
]->data
;
4332 const struct shash_node
**ports
;
4335 if (ofproto
->backer
!= backer
) {
4339 ds_put_format(ds
, "\t%s:\n", ofproto
->up
.name
);
4341 ports
= shash_sort(&ofproto
->up
.port_by_name
);
4342 for (j
= 0; j
< shash_count(&ofproto
->up
.port_by_name
); j
++) {
4343 const struct shash_node
*node
= ports
[j
];
4344 struct ofport
*ofport
= node
->data
;
4346 odp_port_t odp_port
;
4348 ds_put_format(ds
, "\t\t%s %u/", netdev_get_name(ofport
->netdev
),
4351 odp_port
= ofp_port_to_odp_port(ofproto
, ofport
->ofp_port
);
4352 if (odp_port
!= ODPP_NONE
) {
4353 ds_put_format(ds
, "%"PRIu32
":", odp_port
);
4355 ds_put_cstr(ds
, "none:");
4358 ds_put_format(ds
, " (%s", netdev_get_type(ofport
->netdev
));
4361 if (!netdev_get_config(ofport
->netdev
, &config
)) {
4362 const struct smap_node
**nodes
;
4365 nodes
= smap_sort(&config
);
4366 for (i
= 0; i
< smap_count(&config
); i
++) {
4367 const struct smap_node
*node
= nodes
[i
];
4368 ds_put_format(ds
, "%c %s=%s", i
? ',' : ':',
4369 node
->key
, node
->value
);
4373 smap_destroy(&config
);
4375 ds_put_char(ds
, ')');
4376 ds_put_char(ds
, '\n');
4380 shash_destroy(&ofproto_shash
);
4385 ofproto_unixctl_dpif_show(struct unixctl_conn
*conn
, int argc OVS_UNUSED
,
4386 const char *argv
[] OVS_UNUSED
, void *aux OVS_UNUSED
)
4388 struct ds ds
= DS_EMPTY_INITIALIZER
;
4389 const struct shash_node
**backers
;
4392 backers
= shash_sort(&all_dpif_backers
);
4393 for (i
= 0; i
< shash_count(&all_dpif_backers
); i
++) {
4394 dpif_show_backer(backers
[i
]->data
, &ds
);
4398 unixctl_command_reply(conn
, ds_cstr(&ds
));
4403 ofproto_dpif_contains_flow(const struct ofproto_dpif
*ofproto
,
4404 const struct nlattr
*key
, size_t key_len
)
4406 struct ofproto_dpif
*ofp
;
4409 xlate_receive(ofproto
->backer
, NULL
, key
, key_len
, &flow
, &ofp
,
4410 NULL
, NULL
, NULL
, NULL
);
4411 return ofp
== ofproto
;
4415 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn
*conn
,
4416 int argc OVS_UNUSED
, const char *argv
[],
4417 void *aux OVS_UNUSED
)
4419 struct ds ds
= DS_EMPTY_INITIALIZER
;
4420 const struct dpif_flow_stats
*stats
;
4421 const struct ofproto_dpif
*ofproto
;
4422 struct dpif_flow_dump flow_dump
;
4423 const struct nlattr
*actions
;
4424 const struct nlattr
*mask
;
4425 const struct nlattr
*key
;
4429 bool verbosity
= false;
4430 struct dpif_port dpif_port
;
4431 struct dpif_port_dump port_dump
;
4432 struct hmap portno_names
;
4436 ofproto
= ofproto_dpif_lookup(argv
[argc
- 1]);
4438 unixctl_command_reply_error(conn
, "no such bridge");
4442 if (argc
> 2 && !strcmp(argv
[1], "-m")) {
4446 hmap_init(&portno_names
);
4447 DPIF_PORT_FOR_EACH (&dpif_port
, &port_dump
, ofproto
->backer
->dpif
) {
4448 odp_portno_names_set(&portno_names
, dpif_port
.port_no
, dpif_port
.name
);
4452 error
= dpif_flow_dump_start(&flow_dump
, ofproto
->backer
->dpif
);
4456 dpif_flow_dump_state_init(ofproto
->backer
->dpif
, &state
);
4457 while (dpif_flow_dump_next(&flow_dump
, state
, &key
, &key_len
,
4458 &mask
, &mask_len
, &actions
, &actions_len
,
4460 if (!ofproto_dpif_contains_flow(ofproto
, key
, key_len
)) {
4464 odp_flow_format(key
, key_len
, mask
, mask_len
, &portno_names
, &ds
,
4466 ds_put_cstr(&ds
, ", ");
4467 dpif_flow_stats_format(stats
, &ds
);
4468 ds_put_cstr(&ds
, ", actions:");
4469 format_odp_actions(&ds
, actions
, actions_len
);
4470 ds_put_char(&ds
, '\n');
4472 dpif_flow_dump_state_uninit(ofproto
->backer
->dpif
, state
);
4473 error
= dpif_flow_dump_done(&flow_dump
);
4478 ds_put_format(&ds
, "dpif/dump_flows failed: %s", ovs_strerror(errno
));
4479 unixctl_command_reply_error(conn
, ds_cstr(&ds
));
4481 unixctl_command_reply(conn
, ds_cstr(&ds
));
4483 odp_portno_names_destroy(&portno_names
);
4484 hmap_destroy(&portno_names
);
4489 ofproto_dpif_unixctl_init(void)
4491 static bool registered
;
4497 unixctl_command_register(
4499 "{[dp_name] odp_flow | bridge br_flow} [-generate|packet]",
4500 1, 3, ofproto_unixctl_trace
, NULL
);
4501 unixctl_command_register(
4502 "ofproto/trace-packet-out",
4503 "[-consistent] {[dp_name] odp_flow | bridge br_flow} [-generate|packet] actions",
4504 2, 6, ofproto_unixctl_trace_actions
, NULL
);
4505 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
4506 ofproto_unixctl_fdb_flush
, NULL
);
4507 unixctl_command_register("fdb/show", "bridge", 1, 1,
4508 ofproto_unixctl_fdb_show
, NULL
);
4509 unixctl_command_register("dpif/dump-dps", "", 0, 0,
4510 ofproto_unixctl_dpif_dump_dps
, NULL
);
4511 unixctl_command_register("dpif/show", "", 0, 0, ofproto_unixctl_dpif_show
,
4513 unixctl_command_register("dpif/dump-flows", "[-m] bridge", 1, 2,
4514 ofproto_unixctl_dpif_dump_flows
, NULL
);
4518 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
4520 * This is deprecated. It is only for compatibility with broken device drivers
4521 * in old versions of Linux that do not properly support VLANs when VLAN
4522 * devices are not used. When broken device drivers are no longer in
4523 * widespread use, we will delete these interfaces. */
4526 set_realdev(struct ofport
*ofport_
, ofp_port_t realdev_ofp_port
, int vid
)
4528 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(ofport_
->ofproto
);
4529 struct ofport_dpif
*ofport
= ofport_dpif_cast(ofport_
);
4531 if (realdev_ofp_port
== ofport
->realdev_ofp_port
4532 && vid
== ofport
->vlandev_vid
) {
4536 ofproto
->backer
->need_revalidate
= REV_RECONFIGURE
;
4538 if (ofport
->realdev_ofp_port
) {
4541 if (realdev_ofp_port
&& ofport
->bundle
) {
4542 /* vlandevs are enslaved to their realdevs, so they are not allowed to
4543 * themselves be part of a bundle. */
4544 bundle_set(ofport_
->ofproto
, ofport
->bundle
, NULL
);
4547 ofport
->realdev_ofp_port
= realdev_ofp_port
;
4548 ofport
->vlandev_vid
= vid
;
4550 if (realdev_ofp_port
) {
4551 vsp_add(ofport
, realdev_ofp_port
, vid
);
4558 hash_realdev_vid(ofp_port_t realdev_ofp_port
, int vid
)
4560 return hash_2words(ofp_to_u16(realdev_ofp_port
), vid
);
4564 ofproto_has_vlan_splinters(const struct ofproto_dpif
*ofproto
)
4565 OVS_EXCLUDED(ofproto
->vsp_mutex
)
4567 /* hmap_is_empty is thread safe. */
4568 return !hmap_is_empty(&ofproto
->realdev_vid_map
);
4573 vsp_realdev_to_vlandev__(const struct ofproto_dpif
*ofproto
,
4574 ofp_port_t realdev_ofp_port
, ovs_be16 vlan_tci
)
4575 OVS_REQUIRES(ofproto
->vsp_mutex
)
4577 if (!hmap_is_empty(&ofproto
->realdev_vid_map
)) {
4578 int vid
= vlan_tci_to_vid(vlan_tci
);
4579 const struct vlan_splinter
*vsp
;
4581 HMAP_FOR_EACH_WITH_HASH (vsp
, realdev_vid_node
,
4582 hash_realdev_vid(realdev_ofp_port
, vid
),
4583 &ofproto
->realdev_vid_map
) {
4584 if (vsp
->realdev_ofp_port
== realdev_ofp_port
4585 && vsp
->vid
== vid
) {
4586 return vsp
->vlandev_ofp_port
;
4590 return realdev_ofp_port
;
4593 /* Returns the OFP port number of the Linux VLAN device that corresponds to
4594 * 'vlan_tci' on the network device with port number 'realdev_ofp_port' in
4595 * 'struct ofport_dpif'. For example, given 'realdev_ofp_port' of eth0 and
4596 * 'vlan_tci' 9, it would return the port number of eth0.9.
4598 * Unless VLAN splinters are enabled for port 'realdev_ofp_port', this
4599 * function just returns its 'realdev_ofp_port' argument. */
4601 vsp_realdev_to_vlandev(const struct ofproto_dpif
*ofproto
,
4602 ofp_port_t realdev_ofp_port
, ovs_be16 vlan_tci
)
4603 OVS_EXCLUDED(ofproto
->vsp_mutex
)
4607 /* hmap_is_empty is thread safe, see if we can return immediately. */
4608 if (hmap_is_empty(&ofproto
->realdev_vid_map
)) {
4609 return realdev_ofp_port
;
4611 ovs_mutex_lock(&ofproto
->vsp_mutex
);
4612 ret
= vsp_realdev_to_vlandev__(ofproto
, realdev_ofp_port
, vlan_tci
);
4613 ovs_mutex_unlock(&ofproto
->vsp_mutex
);
4617 static struct vlan_splinter
*
4618 vlandev_find(const struct ofproto_dpif
*ofproto
, ofp_port_t vlandev_ofp_port
)
4620 struct vlan_splinter
*vsp
;
4622 HMAP_FOR_EACH_WITH_HASH (vsp
, vlandev_node
,
4623 hash_ofp_port(vlandev_ofp_port
),
4624 &ofproto
->vlandev_map
) {
4625 if (vsp
->vlandev_ofp_port
== vlandev_ofp_port
) {
4633 /* Returns the OpenFlow port number of the "real" device underlying the Linux
4634 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
4635 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
4636 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
4637 * eth0 and store 9 in '*vid'.
4639 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
4640 * VLAN device. Unless VLAN splinters are enabled, this is what this function
4643 vsp_vlandev_to_realdev(const struct ofproto_dpif
*ofproto
,
4644 ofp_port_t vlandev_ofp_port
, int *vid
)
4645 OVS_REQUIRES(ofproto
->vsp_mutex
)
4647 if (!hmap_is_empty(&ofproto
->vlandev_map
)) {
4648 const struct vlan_splinter
*vsp
;
4650 vsp
= vlandev_find(ofproto
, vlandev_ofp_port
);
4655 return vsp
->realdev_ofp_port
;
4661 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
4662 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
4663 * 'flow->in_port' to the "real" device backing the VLAN device, sets
4664 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
4665 * always the case unless VLAN splinters are enabled), returns false without
4666 * making any changes. */
4668 vsp_adjust_flow(const struct ofproto_dpif
*ofproto
, struct flow
*flow
)
4669 OVS_EXCLUDED(ofproto
->vsp_mutex
)
4674 /* hmap_is_empty is thread safe. */
4675 if (hmap_is_empty(&ofproto
->vlandev_map
)) {
4679 ovs_mutex_lock(&ofproto
->vsp_mutex
);
4680 realdev
= vsp_vlandev_to_realdev(ofproto
, flow
->in_port
.ofp_port
, &vid
);
4681 ovs_mutex_unlock(&ofproto
->vsp_mutex
);
4686 /* Cause the flow to be processed as if it came in on the real device with
4687 * the VLAN device's VLAN ID. */
4688 flow
->in_port
.ofp_port
= realdev
;
4689 flow
->vlan_tci
= htons((vid
& VLAN_VID_MASK
) | VLAN_CFI
);
4694 vsp_remove(struct ofport_dpif
*port
)
4696 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
4697 struct vlan_splinter
*vsp
;
4699 ovs_mutex_lock(&ofproto
->vsp_mutex
);
4700 vsp
= vlandev_find(ofproto
, port
->up
.ofp_port
);
4702 hmap_remove(&ofproto
->vlandev_map
, &vsp
->vlandev_node
);
4703 hmap_remove(&ofproto
->realdev_vid_map
, &vsp
->realdev_vid_node
);
4706 port
->realdev_ofp_port
= 0;
4708 VLOG_ERR("missing vlan device record");
4710 ovs_mutex_unlock(&ofproto
->vsp_mutex
);
4714 vsp_add(struct ofport_dpif
*port
, ofp_port_t realdev_ofp_port
, int vid
)
4716 struct ofproto_dpif
*ofproto
= ofproto_dpif_cast(port
->up
.ofproto
);
4718 ovs_mutex_lock(&ofproto
->vsp_mutex
);
4719 if (!vsp_vlandev_to_realdev(ofproto
, port
->up
.ofp_port
, NULL
)
4720 && (vsp_realdev_to_vlandev__(ofproto
, realdev_ofp_port
, htons(vid
))
4721 == realdev_ofp_port
)) {
4722 struct vlan_splinter
*vsp
;
4724 vsp
= xmalloc(sizeof *vsp
);
4725 vsp
->realdev_ofp_port
= realdev_ofp_port
;
4726 vsp
->vlandev_ofp_port
= port
->up
.ofp_port
;
4729 port
->realdev_ofp_port
= realdev_ofp_port
;
4731 hmap_insert(&ofproto
->vlandev_map
, &vsp
->vlandev_node
,
4732 hash_ofp_port(port
->up
.ofp_port
));
4733 hmap_insert(&ofproto
->realdev_vid_map
, &vsp
->realdev_vid_node
,
4734 hash_realdev_vid(realdev_ofp_port
, vid
));
4736 VLOG_ERR("duplicate vlan device record");
4738 ovs_mutex_unlock(&ofproto
->vsp_mutex
);
4742 ofp_port_to_odp_port(const struct ofproto_dpif
*ofproto
, ofp_port_t ofp_port
)
4744 const struct ofport_dpif
*ofport
= get_ofp_port(ofproto
, ofp_port
);
4745 return ofport
? ofport
->odp_port
: ODPP_NONE
;
4748 struct ofport_dpif
*
4749 odp_port_to_ofport(const struct dpif_backer
*backer
, odp_port_t odp_port
)
4751 struct ofport_dpif
*port
;
4753 ovs_rwlock_rdlock(&backer
->odp_to_ofport_lock
);
4754 HMAP_FOR_EACH_IN_BUCKET (port
, odp_port_node
, hash_odp_port(odp_port
),
4755 &backer
->odp_to_ofport_map
) {
4756 if (port
->odp_port
== odp_port
) {
4757 ovs_rwlock_unlock(&backer
->odp_to_ofport_lock
);
4762 ovs_rwlock_unlock(&backer
->odp_to_ofport_lock
);
4767 odp_port_to_ofp_port(const struct ofproto_dpif
*ofproto
, odp_port_t odp_port
)
4769 struct ofport_dpif
*port
;
4771 port
= odp_port_to_ofport(ofproto
->backer
, odp_port
);
4772 if (port
&& &ofproto
->up
== port
->up
.ofproto
) {
4773 return port
->up
.ofp_port
;
4780 ofproto_dpif_alloc_recirc_id(struct ofproto_dpif
*ofproto
)
4782 struct dpif_backer
*backer
= ofproto
->backer
;
4784 return recirc_id_alloc(backer
->rid_pool
);
4788 ofproto_dpif_free_recirc_id(struct ofproto_dpif
*ofproto
, uint32_t recirc_id
)
4790 struct dpif_backer
*backer
= ofproto
->backer
;
4792 recirc_id_free(backer
->rid_pool
, recirc_id
);
4796 ofproto_dpif_add_internal_flow(struct ofproto_dpif
*ofproto
,
4797 struct match
*match
, int priority
,
4798 const struct ofpbuf
*ofpacts
,
4799 struct rule
**rulep
)
4801 struct ofputil_flow_mod fm
;
4802 struct rule_dpif
*rule
;
4806 fm
.priority
= priority
;
4807 fm
.new_cookie
= htonll(0);
4808 fm
.cookie
= htonll(0);
4809 fm
.cookie_mask
= htonll(0);
4810 fm
.modify_cookie
= false;
4811 fm
.table_id
= TBL_INTERNAL
;
4812 fm
.command
= OFPFC_ADD
;
4813 fm
.idle_timeout
= 0;
4814 fm
.hard_timeout
= 0;
4817 fm
.flags
= OFPUTIL_FF_HIDDEN_FIELDS
| OFPUTIL_FF_NO_READONLY
;
4818 fm
.ofpacts
= ofpbuf_data(ofpacts
);
4819 fm
.ofpacts_len
= ofpbuf_size(ofpacts
);
4821 error
= ofproto_flow_mod(&ofproto
->up
, &fm
);
4823 VLOG_ERR_RL(&rl
, "failed to add internal flow (%s)",
4824 ofperr_to_string(error
));
4829 rule
= rule_dpif_lookup_in_table(ofproto
, TBL_INTERNAL
, &match
->flow
,
4840 ofproto_dpif_delete_internal_flow(struct ofproto_dpif
*ofproto
,
4841 struct match
*match
, int priority
)
4843 struct ofputil_flow_mod fm
;
4847 fm
.priority
= priority
;
4848 fm
.new_cookie
= htonll(0);
4849 fm
.cookie
= htonll(0);
4850 fm
.cookie_mask
= htonll(0);
4851 fm
.modify_cookie
= false;
4852 fm
.table_id
= TBL_INTERNAL
;
4853 fm
.flags
= OFPUTIL_FF_HIDDEN_FIELDS
| OFPUTIL_FF_NO_READONLY
;
4854 fm
.command
= OFPFC_DELETE_STRICT
;
4856 error
= ofproto_flow_mod(&ofproto
->up
, &fm
);
4858 VLOG_ERR_RL(&rl
, "failed to delete internal flow (%s)",
4859 ofperr_to_string(error
));
4866 const struct ofproto_class ofproto_dpif_class
= {
4880 NULL
, /* get_memory_usage. */
4881 type_get_memory_usage
,
4900 port_is_lacp_current
,
4901 NULL
, /* rule_choose_table */
4910 rule_modify_actions
,
4924 get_stp_port_status
,
4932 is_mirror_output_bundle
,
4933 forward_bpdu_changed
,
4934 set_mac_table_config
,
4936 NULL
, /* meter_get_features */
4937 NULL
, /* meter_set */
4938 NULL
, /* meter_get */
4939 NULL
, /* meter_del */
4940 group_alloc
, /* group_alloc */
4941 group_construct
, /* group_construct */
4942 group_destruct
, /* group_destruct */
4943 group_dealloc
, /* group_dealloc */
4944 group_modify
, /* group_modify */
4945 group_get_stats
, /* group_get_stats */